WO2010134319A1 - 知識ベースシステム、論理演算方法、プログラム、及び記録媒体 - Google Patents
知識ベースシステム、論理演算方法、プログラム、及び記録媒体 Download PDFInfo
- Publication number
- WO2010134319A1 WO2010134319A1 PCT/JP2010/003327 JP2010003327W WO2010134319A1 WO 2010134319 A1 WO2010134319 A1 WO 2010134319A1 JP 2010003327 W JP2010003327 W JP 2010003327W WO 2010134319 A1 WO2010134319 A1 WO 2010134319A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- attribute
- attributes
- knowledge
- identifier
- true
- Prior art date
Links
- 238000003860 storage Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims description 164
- 238000012545 processing Methods 0.000 claims description 65
- 238000000605 extraction Methods 0.000 claims description 52
- 238000004364 calculation method Methods 0.000 claims description 35
- 238000013500 data storage Methods 0.000 claims description 33
- 235000019640 taste Nutrition 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 11
- 239000003205 fragrance Substances 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 abstract 1
- 238000004891 communication Methods 0.000 description 164
- 210000002569 neuron Anatomy 0.000 description 104
- 230000008569 process Effects 0.000 description 80
- 241000282472 Canis lupus familiaris Species 0.000 description 77
- 230000005540 biological transmission Effects 0.000 description 74
- 238000010586 diagram Methods 0.000 description 65
- 241000124008 Mammalia Species 0.000 description 53
- 230000000875 corresponding effect Effects 0.000 description 45
- 241001465754 Metazoa Species 0.000 description 40
- 238000005516 engineering process Methods 0.000 description 34
- 230000010365 information processing Effects 0.000 description 32
- 241000270322 Lepidosauria Species 0.000 description 25
- 230000006870 function Effects 0.000 description 20
- 241000282414 Homo sapiens Species 0.000 description 19
- 235000002639 sodium chloride Nutrition 0.000 description 18
- FAPWRFPIFSIZLT-UHFFFAOYSA-M Sodium chloride Chemical compound [Na+].[Cl-] FAPWRFPIFSIZLT-UHFFFAOYSA-M 0.000 description 17
- 201000010099 disease Diseases 0.000 description 17
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 17
- 210000000697 sensory organ Anatomy 0.000 description 17
- 239000011780 sodium chloride Substances 0.000 description 17
- 241000282412 Homo Species 0.000 description 16
- 210000004556 brain Anatomy 0.000 description 15
- 210000001508 eye Anatomy 0.000 description 11
- 241000270295 Serpentes Species 0.000 description 10
- 230000010354 integration Effects 0.000 description 10
- 230000007246 mechanism Effects 0.000 description 10
- 241001481833 Coryphaena hippurus Species 0.000 description 8
- 241000282326 Felis catus Species 0.000 description 8
- 239000008280 blood Substances 0.000 description 7
- 210000004369 blood Anatomy 0.000 description 7
- 210000001787 dendrite Anatomy 0.000 description 6
- 210000002741 palatine tonsil Anatomy 0.000 description 6
- 230000000717 retained effect Effects 0.000 description 6
- 239000007787 solid Substances 0.000 description 6
- 208000024891 symptom Diseases 0.000 description 6
- 235000009508 confectionery Nutrition 0.000 description 5
- 206010035664 Pneumonia Diseases 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000003745 diagnosis Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000012217 deletion Methods 0.000 description 3
- 230000037430 deletion Effects 0.000 description 3
- 239000010813 municipal solid waste Substances 0.000 description 3
- 230000008961 swelling Effects 0.000 description 3
- 241000251468 Actinopterygii Species 0.000 description 2
- 235000005979 Citrus limon Nutrition 0.000 description 2
- 244000131522 Citrus pyriformis Species 0.000 description 2
- 241001137251 Corvidae Species 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 210000000887 face Anatomy 0.000 description 2
- 230000001815 facial effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 241001125840 Coryphaenidae Species 0.000 description 1
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- 208000037919 acquired disease Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 235000019658 bitter taste Nutrition 0.000 description 1
- 230000000740 bleeding effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 230000001339 gustatory effect Effects 0.000 description 1
- 239000001257 hydrogen Substances 0.000 description 1
- 229910052739 hydrogen Inorganic materials 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 201000009240 nasopharyngitis Diseases 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 210000003296 saliva Anatomy 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 235000019605 sweet taste sensations Nutrition 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 201000008827 tuberculosis Diseases 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/907—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
Definitions
- the present invention relates to a knowledge base system.
- Things can be expressed in words. In languages like English and Japanese, things are expressed in words, or words. A thing has an attribute which is a fixed property. This attribute is also expressed in words.
- a word can be represented by a character string that is a collection of characters, that is, a word. Therefore, a word representing an object has an attribute, and both an object and its attribute can be represented by a character string written in characters, that is, a word. By using characters, information can be retained and recorded for a long time. Words can also be expressed using words other than letters, such as speech and gestures. The feature is that the word itself represents the meaning of the word itself.
- Objects include not only concrete objects such as stones and the sun, but also abstract objects such as warmth, concepts, images, sounds, and WEB pages. Unknowns may not have attributes.
- the attribute represents the property of the object, such as “has a red color” or “has a sweet taste”.
- a machine such as a computer can compare two character strings and determine whether the two character strings are equal. It is possible to determine whether two character strings are the same, and to generate a true value if they are the same, or a false value if they are different. The computer can determine whether the words represented in letters are the same and generate a truth value.
- a symbol associated with a word represents the meaning of the associated word.
- the symbol in this case represents a meaning.
- the symbol can be said to be an identifier.
- the identifier associated with the word and representing the word represents the meaning of the associated word. Therefore, an identifier associated with a word and representing the meaning of the word is also a word. If the word “dog” is associated with the identifier “ABC”, ABC is a word representing the meaning of the dog.
- Words can be generated not only by character combinations but also by combinations of sounds and figures. Since a word generated by a combination of sounds and figures can be replaced with a word generated by letters, it is not essential what the word is generated except for ease of recording.
- relay type computer There used to be a relay type computer. A mechanism for performing three operations of AND, OR, and NOT using a relay was realized. Although the components of current computers have changed from relays to semiconductors, the principle is the same as for relay computers.
- the current computer does not hold the procedure for performing the logical operation as a combination of machine parts, but holds the procedure for performing the logical operation as information called a program. By changing information called a program, different logical operations can be performed on the same device.
- COBOL is a program that can be understood by humans. Something like COBOL is called a programming language.
- a program that can be directly handled by a computer, which is a machine, is called a machine language.
- a program called a compiler converts a program written in a programming language into a machine language program.
- Objects generally have several attributes. One word representing an object has zero or more attributes.
- PROLOG There is a programming language called PROLOG.
- a word representing an attribute is used to represent the attribute.
- Experiments to realize artificial intelligence using such a programming language were conducted, but there was a limit.
- Socrates is a person in the programming language called PROLOG, it will become (man Associates).
- PROLOG programming language
- the two words “man” and “Socrats” are arranged to express the property of the object, that is, the attribute of the object.
- things and attributes that are properties are stored as words.
- the criterion for determining that “man” is “man” is whether the words are the same.
- the word itself expresses meaning. Words are interpreted as objects, attributes, and relationships depending on the order of the sequence.
- a relationship and two things are stored as a sequence of three words.
- the meaning changes if the order of the word sequence is different. Meaning is included in the order of word arrangement.
- the meaning is expressed by the number and order of words. In this method, a two-level relationship between one object and another can be expressed, but a three-level or higher relationship such as child ⁇ father ⁇ grandfather cannot be expressed directly.
- This technology can hold objects, attributes, and relationships as a database, but it only holds a sequence of words representing things, attributes, and relationships.
- An attribute that is a property of an object is expressed by a word sequence, and a relationship is expressed by a word sequence. Both attributes and relationships are expressed in the same way: word sequences.
- word sequences Assuming that information is systematized, knowledge is stored in a plurality of words that represent information, and the stored word sequence is used as knowledge. In this method, information that directly represents the relationship between objects in three or more layers is not retained.
- the word sequence is associated with each other. By searching for a matching word, the relationship between the information represented by the word sequence is examined each time.
- Patent Document 1 discloses the following technique. Web pages that can be viewed by software called browsers that are stored on a web server on the Internet and run on a personal computer or the like, information such as document files are called nodes, and a group of nodes is called a space. An apparatus for evaluating information held by a node is disclosed. Information linked to a node is called a property. The implementation of properties (attributes) is not specified at all, and the implementation of properties is entirely left up to the program implementer. This technique does not disclose that an object retains an attribute. In addition, an attribute evaluation method is not disclosed.
- Bayesian networks are probabilistic models with a graph structure, to make inferences.
- a certain event it is considered that another event occurs with a certain probability, and a chain in which another event occurs from a certain event is expressed by a graph structure.
- the probability that another event will occur from one event is always maintained.
- An event having the highest occurrence probability can be found by tracing a graph held in advance with a certain event as a starting point.
- inference can be made by this processing.
- things, attributes, and relationships are not distinguished, and all are expressed as events. Events are represented by words in the language, and events are directly connected.
- Bayesian networks are included in the language.
- a meaningful word is generated by combining characters.
- a word is in a language and has the meaning of that word directly in the language. It is a word that has a direct meaning and represents information such as attributes such as things, properties, characteristics, meanings, and concepts.
- a word sequence consisting of a word representing an object and a word representing an attribute is created, and the nature of the object and the relationship between the object and the object are represented by the word sequence.
- a sequence of a plurality of words was accumulated, and the accumulated sequence of words was used as knowledge. In this method, knowledge is built using only words that have the meaning of the word itself.
- the present invention has been made in view of the above problems, and an object of the present invention is to provide a knowledge base system capable of expressing the semantic content of an object or attribute without depending on language.
- a knowledge base system includes a storage unit that stores a knowledge base, and an operation unit that performs a logical operation on the knowledge base stored in the storage unit. It includes an object identifier for identifying an object, and at least one attribute of the object, which is an attribute associated with the object identifier of the object.
- the attribute is associated with an attribute identifier for identifying the attribute, at least one data representing the attribute, feature data associated with the attribute identifier for identifying the attribute, and a word representing the attribute. And at least one of identification data associated with the attribute identifier.
- the object identifier is not a word representing an object, but is composed of a symbol having no meaning in itself.
- the attribute identifier is not a word representing an attribute, but is a symbol having no meaning in itself.
- the object and the attributes of the object can be represented by an identifier that does not represent the meaning of the object or the attribute, retained, and further subjected to calculation.
- a thing is designated means that a thing identifier for identifying the thing is designated.
- the feature data represents at least one of an attribute shape, a sound, an aroma, a taste, a color, a pressure, a temperature, a length, a coordinate value, and an area identified by an attribute identifier associated with the feature data. It may be data. Thus, the concept (meaning content) of each attribute is defined by the feature data associated with the attribute identifier.
- the knowledge base system includes an input unit that acquires information on the attribute, a feature extraction unit that extracts at least one of the feature data and the identification data from the information acquired by the input unit, and the feature A data storage unit that stores at least one of the feature data and the identification data extracted by the extraction unit in the knowledge base in association with an attribute identifier that identifies the attribute may be provided.
- the new relationship between the feature data, the identification data, and the attribute identifier can be stored in the knowledge base.
- the attribute identifier may include a main identifier and a sub-identifier.
- the attribute identifier may identify the attribute by a combination of the main identifier and the sub-identifier.
- the basic identifier can represent the basic property of the attribute
- the sub-identifier can represent a more detailed property such as the degree and appearance.
- the arithmetic unit may perform the logical operation on at least an attribute of the specified object when the one object is specified as an object of the operation.
- the arithmetic unit further provides a true / false value indicating whether the attribute is true or false for each of the attributes. Are collected, only the attributes having the same attribute identifier and whose associated true / false values are both true or false are collected, and the true / false values are associated with individual attributes. If not, an AND operation for generating a new set of attributes may be performed by collecting only attributes having the same attribute identifier.
- the operation unit further collects attributes belonging to at least one of the attribute sets when two attribute sets consisting of at least one attribute are specified, thereby generating a new attribute set.
- An OR operation to be generated may be performed.
- each of the attributes is associated with a true / false value indicating whether the attribute is true or false
- the calculation unit further includes a set of attributes including at least one attribute.
- an attribute that is changed to false is generated if the associated truth value is true, and true if the truth value is false.
- a NOT operation for generating a set of new attributes whose true / false values are changed may be performed by generating the attributes changed to.
- each of the attributes is associated with a true / false value indicating whether the attribute is true or false
- the calculation unit further includes a set of attributes including at least one attribute.
- a set of new attributes is generated by collecting only attributes having the same attribute identifier and whose associated truth values are both true or false. For each attribute included in the generated new attribute set, an attribute that is changed to false is generated if the associated truth value is true, and an attribute that is changed to true if the truth value is false.
- a NAND operation for generating a new set of attributes whose true / false values are changed may be performed.
- each of the attributes is associated with a true / false value indicating whether the attribute is true or false
- the calculation unit further includes a set of attributes including at least one attribute.
- the true / false value is true, an attribute that is changed to false is generated, and if the true / false value is false, an attribute that is changed to true is generated. You may perform the NOR calculation to produce
- the arithmetic unit when the two sets of attributes including at least one attribute are specified, the arithmetic unit further indicates whether each attribute is true or false indicating whether the attribute is true or false.
- the number of attributes that are included in any set of attributes and that have the same associated truth value is counted as a common degree, and each of the attributes is counted as the authenticity.
- the number of attributes included in any attribute set may be counted as the degree of commonality.
- the arithmetic unit when the two sets of attributes including at least one attribute are specified, the arithmetic unit further indicates whether each attribute is true or false indicating whether the attribute is true or false.
- the number of attributes that are included in any set of attributes and that have different associated truth values is counted as a degree of dissimilarity. If they are not associated with each other, the number of attributes included only in one of the attribute sets may be counted as the non-commonness.
- the attribute identifier may include a main identifier and a sub-identifier. Further, in the case where two sets of attributes including at least one attribute are designated, the arithmetic unit is associated with each attribute with a true / false value indicating whether the attribute is true or false. If the number of attributes having the same primary identifier and the same true / false value is counted as a similarity, and the true / false value is not associated with each of the attributes, The number of attributes having the same main identifier may be counted as the similarity.
- the attribute identifier may include a main identifier and a sub-identifier. Further, in the case where two sets of attributes including at least one attribute are designated, the arithmetic unit is associated with each attribute with a true / false value indicating whether the attribute is true or false. When the number of attributes having the same primary identifier and different true / false values is counted as dissimilarity, and the true / false values are not associated with individual attributes, The number of attributes included only in the set may be counted as dissimilarity.
- each of the attributes is associated with a true / false value indicating whether the attribute is true or false
- the calculation unit further includes a set of attributes including at least one attribute.
- an attribute that belongs to at least one of the attribute sets and that is included in any attribute set and that has a different associated truth value is excluded.
- An OR operation for generating a set of attributes may be performed.
- the arithmetic unit further generates a new attribute set by excluding attributes belonging to any attribute set when two attribute sets consisting of at least one attribute are specified. XOR operation may be performed.
- the knowledge base system can execute various arithmetic processes using a specified set of attributes as an operation target.
- the “attribute set” typically refers to a set of zero or more attribute identifiers associated with the object identifier, but is not limited thereto.
- Cold attributes refers to collecting attribute identifiers that meet a condition.
- the calculation unit may generate a collection of knowledge having the attribute by collecting object identifiers associated with the specified attribute.
- the object it may be determined whether the object identifier of the object belongs to the knowledge collection.
- the arithmetic unit includes an object identifier for identifying the generated collection of knowledge, an attribute identifier for identifying the specified at least one attribute, and an attribute identifier for an attribute indicating an abstract object.
- An abstract object may be newly generated by association.
- the knowledge base may include a knowledge network that is a connection of nodes associated with objects.
- the node may include an object identifier corresponding to the node and knowledge network information that is information about the knowledge network to which the node belongs.
- the knowledge network information may include a knowledge network identifier for identifying the knowledge network and a pointer to another node connected to the node in the knowledge network.
- the arithmetic unit is connected to the specified object by referring to the object identifier and knowledge network information included in the nodes constituting the knowledge network. You may identify other things that you have. Furthermore, when the attribute identifier is designated, the arithmetic unit may search a knowledge network associated with the attribute identifier and obtain a knowledge network identifier of the searched knowledge network. As a result, one thing can be used as a starting point to reach other things that may be related to the thing. This process is associated with “inference”.
- the arithmetic unit further associates an entity identifier for identifying the knowledge network with at least one attribute identifier associated with the knowledge network and an attribute identifier for an attribute indicating an abstract entity.
- An abstract object may be newly generated.
- the logical operation method includes a step of performing processing by at least the above-described arithmetic unit on the knowledge base included in the above-described knowledge base system.
- a program according to an embodiment of the present invention is a program for a knowledge base system, and causes a computer to execute the steps included in the logical operation method described above.
- a recording medium is a computer-readable recording medium for a knowledge base system, in which the knowledge base described above and the program described above are recorded.
- the present invention can be realized not only as a knowledge base system but also as a program for causing a computer to execute the functions of the knowledge base system. Needless to say, such a program can be distributed via a recording medium such as a CD-ROM and a transmission medium such as the Internet.
- FIG. 1 is a diagram for comparing and explaining a conventional knowledge base and a knowledge base according to an embodiment of the present invention.
- FIG. 2A is a diagram showing a first-layer graphic representing the relationship between attributes and feature data.
- FIG. 2B is a diagram showing a first-layer graphic representing the relationship between attributes and identification data.
- FIG. 3 is a diagram showing a graphic of the second hierarchy representing the relationship between objects and attributes.
- FIG. 4 is a diagram illustrating a graphic of the third hierarchy representing the relationship between objects.
- FIG. 5 is a diagram showing a figure in which the first to third layers are integrated.
- FIG. 6 is a conceptual diagram of a knowledge base system according to an embodiment.
- FIG. 7 is a block diagram of a computer system according to an embodiment.
- FIG. 1 is a diagram for comparing and explaining a conventional knowledge base and a knowledge base according to an embodiment of the present invention.
- FIG. 2A is a diagram showing a first-layer graphic representing the relationship between attributes and feature data
- FIG. 8 is a functional block diagram of a knowledge base system according to an embodiment.
- FIG. 9 is a diagram showing a correspondence table between attribute identifiers and feature data according to an embodiment.
- FIG. 10 is a diagram illustrating a correspondence table between an object identifier and an attribute identifier according to an embodiment.
- FIG. 11 is a diagram showing two-dimensional orthogonal coordinates.
- FIG. 12 is a diagram showing three-dimensional coordinates that intersect at an angle smaller than a right angle.
- FIG. 13 is a diagram showing an example in which attributes are represented by coordinate axes.
- FIG. 14 is a diagram illustrating an example of a fuzzy set.
- FIG. 15 is a diagram illustrating the relationship between objects.
- FIG. 16 is a diagram illustrating a relationship between a knowledge network and attributes.
- FIG. 17A is a diagram illustrating a data list that is a source of attributes.
- FIG. 17B is a diagram showing a correspondence table between object names and object identifiers.
- FIG. 18 is a diagram showing a correspondence table between attributes and attribute identifiers.
- FIG. 19 is a diagram showing a list of objects and attributes.
- FIG. 20A is a diagram showing a correspondence table of objects, main attributes, and sub-attributes.
- FIG. 20B is a diagram showing a correspondence table between main attributes and sub-attributes.
- FIG. 21 is a flowchart illustrating an AND operation.
- FIG. 22 is a flowchart illustrating the OR operation.
- FIG. 23 is a flowchart illustrating the NOT calculation.
- FIG. 21 is a flowchart illustrating an AND operation.
- FIG. 22 is a flowchart illustrating the OR operation.
- FIG. 23 is a flowchart illustrating the NOT calculation.
- FIG. 21 is a flowchart
- FIG. 24 is a diagram showing a flowchart of the NAND operation.
- FIG. 25 is a flowchart showing the XOR operation.
- FIG. 26 is a flowchart illustrating a process for calculating the commonality.
- FIG. 27 is a diagram illustrating a flowchart of a process for calculating the dissimilarity.
- FIG. 28A is a diagram showing a correspondence table between object names and object identifiers.
- FIG. 28B is a diagram showing a correspondence table between knowledge networks and attributes.
- FIG. 28C is a diagram illustrating a data structure of the knowledge network.
- FIG. 29 is a diagram illustrating a data structure of a node.
- FIG. 30 is a diagram showing attribute hierarchy information.
- FIG. 31 is a flowchart of processing for searching a knowledge collection.
- FIG. 31 is a flowchart of processing for searching a knowledge collection.
- FIG. 32 is a flowchart of processing for tracing the knowledge network.
- FIG. 33 is a diagram illustrating an example of a data structure in the knowledge base.
- FIG. 34 is a diagram illustrating an example of a data structure in the knowledge base.
- FIG. 35 is a diagram illustrating an example of a data structure in the knowledge base.
- FIG. 36 is a diagram illustrating an example of a data structure in the knowledge base.
- FIG. 37 is a diagram for explaining a knowledge collection in the knowledge base.
- FIG. 38 is a diagram for explaining a knowledge collection in the knowledge base.
- FIG. 39 is a diagram for explaining a knowledge network in the knowledge base.
- FIG. 40 is a diagram for explaining a knowledge network in the knowledge base.
- FIG. 41 is a diagram for explaining a knowledge network in the knowledge base.
- FIG. 42 is a diagram illustrating an example of a knowledge network.
- FIG. 43 is a diagram illustrating the relationship among objects, attributes, and feature data.
- FIG. 44 is a diagram illustrating an example of a knowledge network.
- FIG. 45 is a diagram illustrating an example of a knowledge network.
- FIG. 46 is a diagram showing a triangle.
- FIG. 47 is a diagram illustrating a knowledge network in which a triangle is represented by three points.
- FIG. 48 is a diagram showing a knowledge network in which a triangle is represented by three points and three lines.
- FIG. 49 is a diagram showing a knowledge network in which triangles are represented by three lines.
- FIG. 50 is a diagram illustrating a triangular pyramid.
- FIG. 51 is a diagram illustrating a knowledge network representing a triangular pyramid.
- FIG. 52 is a schematic view of the eyes, nose, and mouth.
- FIG. 53 is a diagram showing a knowledge network representing the distance between eyes, nose and mouth.
- FIG. 54 is a diagram showing a knowledge network representing a shogi board.
- FIG. 55 is a diagram showing an example of implementation of the correspondence table by the network.
- FIG. 56 is a diagram showing a correspondence table between feature data and attributes.
- FIG. 57 is a diagram showing a correspondence table between one feature data and an attribute.
- FIG. 58 is a diagram showing a correspondence table between one attribute and feature data.
- Fig. 1 The figure on the left side of Fig. 1 is the knowledge base of the prior art.
- the knowledge base of the prior art is included in languages such as English, Japanese, and Chinese. Since the knowledge base of the prior art holds words in the language, there is always a portion that overlaps the language. The knowledge base of the prior art loses the common part with the language and cannot be generated outside the language.
- the objects in the real world are directly connected to words in the language.
- the object is not a word.
- the knowledge base is composed of words in a language.
- Prior art knowledge bases are included in the language.
- the word “dog” is directly associated with the real world object dog that is the source of the data.
- the word “dog” not only identifies the dog that is the object, but also directly includes the nature, characteristics, meaning, or concept of the dog that is the object. Only the word “dog” can be used to understand the characteristics, nature, meaning, or concept of a mammal, four-legged dog. Words exist only in the language, and the words themselves directly contain their properties, characteristics, meanings or concepts. Thus, a word in a language directly represents the nature, characteristics, meaning or concept of the word with just that word. The real world and language have no common part. And there is language outside the real world. The object is only in the real world. Words and words are only in the language. A word or word directly identifies an object or an attribute of the object. Furthermore, a word or word directly includes the meaning of the word or word itself.
- This knowledge base is not included in the language and has no common parts with the language. This knowledge base is outside the language. This knowledge base is outside the real world. Furthermore, this knowledge base has nothing in common with the real world.
- feature data generated from information such as images and sounds directly emitted from objects in the real world are stored and held in the knowledge base. For example, feature data (image data) that is emitted directly from a dog of an object that is not a word and generated from the acquired video of the dog is stored and held in the knowledge base.
- identification data is generated from the word by the feature extraction unit, and stored and held in the knowledge base. Identification data is not generated from information that is directly emitted from an object that is not a word, but is generated from information that is generated from a word in a language.
- the feature data is not generated from words in the language but is generated from information directly emitted from the object.
- Feature data is generated from information such as video and sound directly emitted from objects in the real world and stored in the knowledge base.
- the object is not a word.
- an attribute is generated in the knowledge base, and the stored feature data is connected to the attribute.
- objects are generated in the knowledge base. Receives direct information from non-word objects, generates input data from the received information, generates feature data from the input data, connects the generated feature data and attributes, and generates the attributes Connected with things.
- attributes connected to feature data and connections to objects are created by a knowledge base system.
- feature data are not directly connected. Even among feature data connected to the same attribute, each feature data is once connected to an attribute, and is connected to other feature data via the attribute.
- the identification data is the same, and the identification data is not directly connected. Even among identification data connected to the same attribute, each identification data is once connected to the attribute, and is connected to other identification data via the connected attribute. Feature data and identification data are not directly connected. Even in the feature data and the identification data connected to the same attribute, the feature data is once connected to the attribute and connected to the identification data through the connected attribute.
- the attribute and the feature data are connected without going through the object, other attributes, identification data, and other feature data.
- Such a connection between the attribute and the feature data is described as a direct connection between the attribute and the feature data.
- the attribute and the identification data are connected without going through an object, other attributes, other identification data, and feature data.
- Such a connection between the attribute and the identification data is described as a direct connection between the attribute and the identification data.
- an object and an attribute are connected without passing through another object, another attribute, identification data, and feature data.
- Such a connection between an object and an attribute is described as a direct connection between the object and the attribute.
- things are connected to each other without passing through other things, attributes, identification data, and feature data.
- This way of connecting things is described as connecting things directly.
- an object and identification data may be connected without passing through another object, attribute, other identification data, and feature data.
- Such a connection between the object and the identification data is described as a direct connection between the object and the identification data.
- an object and feature data may be connected without passing through another object, attribute, identification data, and other feature data. This way of connecting the object and the feature data is described as connecting the object and the feature data directly.
- the attribute is identified by the identification data by being directly connected to the identification data. If an attribute that is directly connected to identification data is connected to an object, an attribute that is directly connected to identification data can be acquired by the identification data, and an object that is directly connected to the acquired attribute can be acquired. With the identification data, only an attribute directly connected to the identification data and an object directly connected to the attribute can be acquired.
- the identification data is used only to associate a language outside the knowledge base of one embodiment of the present technology with objects and attributes in the knowledge base of one embodiment of the present technology.
- the attribute identifier and the identification data are common in that both are attribute identification information and information that supports identification by a computer (knowledge base system).
- the attribute identifier is information (data) associated with the attribute
- the identification data is different in that it is a word or information (data) associated with the word.
- the feature data and the identification data are common in that they are specific examples (instances) of attributes.
- feature data is an attribute entity
- identification data is different in that a computer (knowledge base system) is data for identifying an attribute.
- the word and the word are information (data) that supports identification by a human.
- a word and a conceptual collection of words is a “language” or “language world”.
- a figure in which only an object and an attribute are directly connected and associated with each other is a figure in the second hierarchy shown in FIG. In this hierarchy, zero or more attributes are directly associated with one thing.
- the number of attributes directly associated with one object is not limited to one. In this hierarchy, things are not connected.
- a figure in which only objects are directly connected and associated is the figure in the third hierarchy shown in FIG. In this hierarchy, things and attributes are not connected.
- the connection only between these things can be various ways such as straight line, ring, tree structure, network structure.
- a figure in which only attributes and feature data are directly connected and associated is the figure in the first hierarchy shown in FIG. 2A.
- the attribute itself has no direct meaning.
- the attribute has a meaning or the meaning of the attribute is generated.
- a concept is generated by directly connecting attributes and feature data. For example, the feature data generated from the video of the shirt turned inside out is directly connected to the attribute, and the meaning or concept of “inside out” is generated.
- One attribute may be directly connected to multiple feature data.
- one meaning or one concept can be generated from a plurality of feature data.
- the attribute representing the Romanesque style is directly connected to a plurality of feature data generated from videos of a plurality of buildings in the Romanesque style.
- the number of feature data directly associated with one attribute is not limited to one.
- One feature data may be directly connected to multiple attributes.
- the attribute directly associated with one feature data is not limited to one.
- the feature data of the bonfire video is directly linked to an attribute, and the concept of bonfire color is generated.
- the feature data of the bonfire video is directly connected to another attribute, and the concept of bonfire warmth is generated.
- the thing itself has no meaning.
- the nature, characteristics, meaning, or concept of an object is determined by the attributes that are directly connected to the object and associated with each other.
- FIG. 5 is a diagram in which three layers are connected.
- one attribute corresponds to one nerve cell
- one object corresponds to one nerve cell.
- the feature extraction unit in the brain extracts features from the received information, and stores the extracted features as feature data in a storage area in the brain.
- Various features such as shape, incense, and taste are stored as feature data.
- a nerve cell that is an attribute of an attribute region in the brain is connected to feature data, and a meaning or concept is generated.
- An attribute is connected to a nerve cell that is an object in the brain, so that the property, characteristic, meaning, or concept of the object is determined.
- One feature data may generate multiple concepts. Assume that the feature data of the flame image is in the knowledge base. This feature data is connected to a certain attribute and generates the concept of flame color. Also, the same feature data is connected to another attribute and generates the concept of warmth. You may feel warmth just by looking at the picture of the fire burning in the fireplace.
- One attribute is connected to multiple feature data. Assume that there is an attribute connected to the characteristic data of acidity.
- the feature data of the image of the lemon is connected to the sour attribute, and the concept of sourness is generated.
- the characteristic data of the blue apple image may be connected to the above-mentioned attribute of sourness. Saliva may be produced by just looking at the lemon, but it can be explained by thinking like this. In this way, different feature data may be connected to one and the same attribute to generate one concept.
- the above-mentioned three-layer structure can express and store knowledge without using identification data generated from words, and systematize the knowledge. Using this method, even animals without words can systematically maintain knowledge and process knowledge such as inference.
- identification data generated from the video of words represented by letters or the sound of words represented by speech, etc. are retained, and the identification data and attributes can be linked and correlated. it can. If the feature data is directly connected to the attribute that is directly connected to the identification data, the word in the language from which the identification data is based and the feature data connected to the attribute are indirect via the identification data and the attribute. Are associated with each other. If the feature data is not directly connected to the attribute directly connected to the identification data and there is an object directly connected to the attribute, the word in the language from which the identification data is based and the object directly connected to the attribute , The identification data and the attribute are indirectly associated with each other. Of course, the identification data can be directly connected to one object.
- one attribute represents a plurality of identification data in a bundle. Is more convenient. If the identification data generated from the video of the character is directly connected to the attribute directly connected to the identification data generated from the audio, the identification data generated from the video of the character is immediately acquired without acquiring other attributes. You can get it. Quick conversion from voice to text. For example, it is possible to acquire an attribute directly connected to identification data generated from speech, and then acquire identification data generated from a video of a character connected to the attribute.
- the identification data associates the word in the language that is the basis of the identification data with the object or attribute in the knowledge base. However, it merely associates the identification data of words represented by video or audio with attributes and objects. Identification data does not directly represent or generate properties, features, meanings or concepts, but only identifies attributes and objects in the knowledge base.
- Information received directly from the object itself is received using a sensory organ, and feature data extracted from the received data is the source of properties, features, meanings or concepts. It is also possible to directly connect identification data generated from words and objects.
- the object is directly identified by them.
- the feature data or identification data may be directly connected to the object.
- the feature data or identification data directly connected to the object is not limited to one, and a plurality of feature data or identification data may be directly connected to the object.
- Feature data or identification data directly connected to an object can be directly connected to an attribute at the same time, and the feature data can generate a meaning and a concept.
- Specify feature data and attributes issue instructions to the learning unit of the knowledge base system from outside the knowledge base system to connect the specified feature data and attributes, and specify the feature data and specified attributes specified in the learning unit Can be connected to learn.
- the human brain may work instead of the part where other parts of the brain do not work even if part of the brain does not work due to internal bleeding. This can be explained by thinking that other nerve cells play the same role by connecting other nerve cells to things instead of nerve cells that have stopped working.
- Objects and attributes correspond to nerve cells. Objects and attributes themselves have no meaning, and objects and attributes are simply identified by identifiers that have no meaning for objects or attributes. Even if you switch to another object or attribute, you can show the same correspondence as before switching.
- All objects can be identified with a single identifier.
- a single form of identifier can identify all attributes.
- All feature data and identification data can be identified with a single type of identifier.
- a single type of identifier can also identify all things, all attributes, all feature data and identification data. Two types of identifiers or three types of identifiers may be used. It is possible to use more than four types of identifiers, but the mechanism is only complicated.
- Both identification data and feature data can be connected to one attribute. It is also possible to create only two types of attributes, that is, an attribute connected only to identification data and an attribute connected only to feature data.
- an attribute connected only to identification data and an attribute connected only to feature data are connected to one object. For example, if there is identification data in the knowledge base that matches the speech identification data generated from the speech of a word heard by the ear, the attribute directly connected to the matching feature data is acquired, and the object directly connected to the attribute is obtained. get. Furthermore, if other attributes directly connected to the object are acquired and feature data directly connected to the acquired attribute is acquired, the nature, characteristics, meaning, or concept of the object represented by the word can be known. This is a function corresponding to association. Further, identification data is generated from one word, and the generated identification data is directly connected to one attribute, and when the attribute is directly connected to several different objects, different objects can be represented by one identification data. This corresponds to an ambiguous word.
- Identification data and feature data can be directly connected to the same attribute.
- an attribute directly connected to the identification data is acquired, and feature data directly connected to the acquired attribute is acquired.
- the acquired feature data represents the nature, feature, meaning or concept directly included in the word from which the identification data is based.
- identification data generated from one word is directly connected to a plurality of different attributes
- the identification data generated from one word is associated with a plurality of meanings or concepts. This also corresponds to an ambiguous word. If the identification data is directly connected to the object, the object can be directly selected and identified by the identification data. Since an object can be acquired without acquiring an attribute, an object can be acquired more quickly.
- knowledge can be accumulated and knowledge processing such as recognition and inference can be performed without identification data.
- the fourth information process is positioned between the second information process and the third information process.
- the knowledge base system according to the embodiment of the present technology can perform the fourth information processing.
- Knowledge can be accumulated and processed using a knowledge base consisting only of objects, attributes, and identification data.
- the knowledge base system in this case does not include properties, features, meanings or concepts. Since the object and the attribute are identified only by the identification data and the feature data is not processed, the object and the attribute can be processed more quickly and the processing load can be reduced. Even if the knowledge base includes feature data, it is possible to perform knowledge processing such as inference using only objects, attributes, and identification data.
- a crow that feeds the city center of Tokyo breaks the garbage bags that were put out late at night or early in the morning and treats them with food.
- Garbage is scattered from the bag, detracting from aesthetics. Humans take measures by changing the color of the bag to make it difficult to see the contents, or by adding ingredients that hate crows to the bag.
- the crow continues to beat the trash in the bag overcoming the measures taken by humans. It seems that crows are learning and comparing wisdom with humans. It is unlikely that learning as seen in the crow example is performed in the third information processing. It is natural to think that learning is performed in a simpler method than the third information processing. In the present application, this is called fourth information processing.
- fourth information processing When knowledge expressed in the above-described three-layer structure is constructed, fourth information processing can be performed.
- FIG. 6 shows a knowledge base system 1000 which is an embodiment of the present technology.
- a camera, a microphone, an infrared sensor, and the like are connected to the computer as input devices.
- the computer shown in FIG. 7 is used in the knowledge base system 1000 shown in FIG.
- FIG. 7 is a conceptual diagram of a computer.
- the computer shown in FIG. 7 includes an input device 1, an arithmetic device (Central Processing Unit: CPU) 2, a communication device 3, an external storage device 5, a main storage device 6, and an auxiliary storage device 7.
- CPU Central Processing Unit
- the arithmetic device 2 issues an instruction to the auxiliary storage device 7 and reads the program 9 stored in the auxiliary storage device 7 and the data 10 used by the program 9 into the main storage device 6. Further, the program and data in the main storage device 6 are stored in the auxiliary storage device 7 in accordance with an instruction from the arithmetic device 2. The arithmetic device 2 issues an instruction to the external storage device 5 and reads the data 12 and the program 11 stored in the external storage device 5 into the main storage device 6. Further, the program and data in the main storage device 6 are stored in the external storage device 5 in accordance with an instruction from the arithmetic device 2.
- the input device 1 includes a keyboard, a mouse, a scanner for reading an image, a camera, a microphone, a temperature sensor, a pressure sensor, and the like. Analog data such as images, sounds, and temperatures are converted into digital data.
- the output device 8 includes a display device such as a flat display, a printer, a device that outputs sound (speaker), a relay contact, a device that controls equipment, and the like.
- the computing device 2 gives an instruction to the communication device 3 and communicates with other computers and devices not shown through the network 4.
- the arithmetic device 2 reads the program 9 of the auxiliary storage device 7 or the program 11 of the external storage device 5 into the main storage device 6, the arithmetic device 2 performs processing according to the procedure described in the read programs 9 and 11.
- the programs 9 and 11 include input from the input device 1, output to the output device 8, communication using the communication device 3, reading and writing of data with the main storage device 6, auxiliary storage device 7 and external storage A processing procedure for reading and writing data with the device 5 is included.
- the knowledge base program 13 is a program that performs logical operations, knowledge collections, and knowledge network processing using the present technology.
- the relational database 14 creates tabular data 10 and 12 in the auxiliary storage device 7 or the external storage device 5, and performs processing of adding / searching / referring / updating / deleting data to the created table.
- the knowledge base program 13 uses the relational database 14 to add, search, reference, update, and delete data from the table.
- FIG. 8 is a functional block diagram of the knowledge base system 1000 according to an embodiment of the present invention.
- Each functional block shown in FIG. 8 is stored as the knowledge base program 13 in the auxiliary storage device 7 shown in FIG. 4, for example.
- the knowledge base program 13 includes, for example, a data input unit 1010, a feature extraction unit 1020, a data comparison unit 1030, a data storage unit 1040, a learning unit 1050,
- a program 9 is read into the main storage device 6 and executed by the arithmetic device 2.
- the data input unit 1010 and the output unit 1080 are also read into the main storage device 6 and executed by the arithmetic device 2.
- the data input unit 1010 automatically acquires video data with a camera and acquires sound data with a microphone as follows. For example, when it is determined by an infrared sensor that a person or an animal is in front of the camera, this triggers acquisition of video data with the camera and acquisition of sound data with a microphone.
- video data may be acquired with a camera and sound data may be acquired with a microphone due to a human clicking operation.
- Feature extraction unit 1020 generates feature data or identification data from the input data input to data input unit 1010.
- the data comparison unit 1030 includes feature data (or identification data) extracted by the feature extraction unit 1020 and feature data (or identification data) already stored in the data 10 (or data 12, the same applies hereinafter). And compare.
- the data storage unit 1040 stores the new feature data or identification data in the data 10. Further, the data storage unit 1040 generates an attribute identifier of a new attribute and stores it in the data 10 in association with the new feature data or identification data. Further, the data storage unit 1040 generates an object identifier of a new object and stores it in the data 10 in association with the attribute identifier of the new attribute.
- the data storage unit 1040 actually requests the relational database 14 to store new feature data or identification data in the data 10.
- the relational database 14 is requested to perform processing such as data addition, search, reference, update, and deletion, and the processing result is obtained. Obtained from the relational database 14.
- the description regarding the processing of the relational database 14 is omitted for the sake of simplicity.
- the data input unit 1010 acquires input data such as video data and sound data directly emitted from the object.
- the input data such as video data or sound data acquired by the data input unit 1010 is extracted by the feature extraction unit 1020 and becomes feature data or identification data.
- the feature data or the identification data is stored in a knowledge base in the data 10 after being subjected to several processes.
- the feature extraction unit 1020 extracts feature data from the input data acquired from the data input unit 1010 and passes the extracted feature data to the data comparison unit 1030.
- the data comparison unit 1030 compares the feature data already stored in the data 10 with the passed feature data and checks whether there is the same feature data.
- the data storage unit 1040 stores the feature data in the knowledge base, generates an attribute in the knowledge base, and directly connects the stored feature data and the generated attribute. Further, the data storage unit 1040 generates an object, and directly connects the generated object to the generated attribute.
- the data storage unit 1040 stores the generated identification data in the knowledge base, generates an attribute in the knowledge base, and directly connects the stored identification data and the generated attribute. . Further, the data storage unit 1040 generates an object, and directly connects the generated object to the generated attribute.
- One attribute directly connected to some feature data represents all the connected feature data by bundling and consolidating all the connected feature data, and represents the nature, feature, meaning or concept.
- One attribute connected to only one feature data also represents the nature, feature, meaning or concept of the feature data.
- the feature data connected to the attribute is not limited to one.
- One attribute that is directly connected to some identification data represents all connected identification data by bundling and consolidating all connected identification data.
- the identification data connected to the attribute is not limited to one.
- FIG. 9 is a diagram illustrating an example of a correspondence table between attribute identifiers and feature data.
- FIG. 10 is a diagram illustrating an example of a correspondence table between object identifiers and attribute identifiers.
- the attribute identifier and the feature data identifier are held in association with each other. That is, the “attribute” in this specification includes an attribute identifier and feature data associated with the attribute identifier, as shown in FIG.
- the attribute identifier is not a word representing the attribute itself, but is composed of a symbol (in this example, a numeric string) that has no meaning in itself.
- the feature data is data representing the semantic content of the attribute, for example, the shape, sound, incense, taste, color, pressure, temperature, length, coordinate value, and area of the attribute identified by the attribute identifier This data represents at least one of the following.
- FIG. 9 may hold an attribute identifier and identification data (or its identifier) associated with the attribute identifier.
- the identification data is data generated from words representing the attribute.
- a plurality of feature data may be associated with one attribute identifier, and one feature data (or identification data) has different attributes. Sometimes associated with an identifier.
- the object identifier and the attribute identifier are stored in association with each other.
- the object identifier is not a word representing the object itself, but is composed of a symbol having no meaning in itself. That is, “thing” in the present specification is represented by a thing identifier and a set of zero or more attribute identifiers associated with the thing identifier.
- the data storage unit 1040 stores new feature data (or identification data; the same applies hereinafter) extracted by the feature extraction unit 1020 in the data 10, and then generates a new attribute identifier. At this time, the data storage unit 1040 generates a new attribute identifier so as not to overlap with the attribute identifier already registered in FIG. Then, the data storage unit 1040 associates the newly generated attribute identifier with the identifier of the feature data stored in the data 10 and registers it in the attribute identifier / feature data correspondence table shown in FIG.
- the data storage unit 1040 generates a new object identifier.
- the data storage unit 1040 generates a new object identifier so as not to overlap with the object identifier already registered in FIG.
- the data storage part 1040 matches the said object identifier and the above-mentioned attribute identifier, and registers them in the correspondence table of the object identifier and attribute identifier shown by FIG.
- an unused object identifier may be held in the data 10, and a new object identifier to be used may be acquired from the unused object identifiers. Further, an unused attribute identifier may be held in the data 10 and an attribute identifier to be newly used may be acquired from the unused attribute identifier. Further, an identifier of unused feature data or an identifier of unused identification data is held in the data 10, and is newly used from the identifier of unused feature data or the identifier of unused identification data. An identifier of feature data or an identifier of identification data may be acquired. If the attribute identifier, object identifier, feature data identifier, and identification data identifier are in the same common format, the mechanism becomes simple.
- the learning unit 1050 arranges the relationship between the object identifier, the attribute identifier, and the feature data identifier (identification data identifier) shown in FIGS. 9 and 10 based on information acquired from the outside.
- the search unit 1060 searches the correspondence tables shown in FIGS. 9 and 10 in accordance with instructions from the data comparison unit 1030 or the learning unit 1050. Details of these processes will be described later.
- the operation unit 1070 performs a logical operation on the knowledge base stored in the data 10. More specifically, when at least one object identifier is designated as an operation target, a logical operation is performed on an attribute associated with the object identifier. Specific examples of the logical operation include, for example, AND operation, OR operation, NOT operation, NAND operation, NOR operation, common degree calculation, non-common degree calculation, similarity calculation, and dissimilarity calculation. . Details of these processes will be described later.
- the output unit 1080 functions as a device driver that outputs the learning result of the learning unit 1050, the calculation result of the calculation unit 1070, and the like to the output device 8.
- an object identifier for identifying the object and an attribute identifier for identifying the attribute are processing targets.
- the thing and the attribute are connected means that the object identifier and the attribute identifier are associated with each other and registered in the correspondence table of the object identifier and the attribute identifier shown in FIG.
- “acquiring an object connected to an attribute” indicates that an object identifier associated with the attribute identifier is acquired.
- “generate an object” refers to generating a new object identifier. The above example is an example, and other similar descriptions shall be similarly interpreted.
- the knowledge base system is an information system that handles knowledge
- the information format of the knowledge base system is also a knowledge format.
- the basic unit of knowledge is a thing.
- the objects include not only concrete objects such as stones and mountains, but also abstract objects such as cold, size, WEB page, and sound.
- Objects can have attributes that are predetermined properties.
- the attribute of one object is not limited to one, and can have zero or more attributes.
- a collection of zero or more attributes is called a set of attributes.
- the logical operation has an aspect of operation on a set. When performing logical operations on attributes, it may be convenient to consider the entire attribute of one object as a set of attributes.
- One attribute set has zero or more attributes. Things have attributes as attached information. Alternatively, an object has a set of attributes as attached information.
- the thing having an attribute includes not only having the attribute itself but also having a pointer to the attribute. Similarly, having an attribute set includes having a pointer to the attribute set in addition to having the attribute set itself.
- the X-axis and the Y-axis intersect at right angles, and a figure on a plane can be expressed. If the three coordinates of the X axis, the Y axis, and the Z axis intersect at right angles, a three-dimensional solid can be represented. If the X axis, Y axis, and Z axis intersect at an angle smaller than a right angle, the result is as shown in FIG. 12, and three coordinates can be expressed in a plane. If the angle at which the coordinate axes intersect is further reduced, more coordinate axes can be represented by a plane.
- FIG. 13 is an example in which attributes are associated with coordinate axes in one embodiment.
- the attribute is expressed by a three-dimensional coordinate axis.
- One coordinate axis represents one attribute.
- the coordinate axis 1 expresses the attribute “high”
- the coordinate axis 2 expresses the attribute “salty”
- the coordinate axis 3 expresses the attribute “sweet”.
- the attributes are only associated with the coordinate axes for easy explanation. Corresponding attributes to coordinate axes is not essential for this embodiment. Further, in order to make the explanation of the present embodiment easy to understand, in FIG. 13, the coordinate axes 1, 2, and 3 are merely orthogonal. The angle at which the coordinate axes intersect is not essential to the present embodiment. It is an example for facilitating the explanation.
- attributes are associated with coordinate axes.
- the attribute is represented by a pair of a coordinate value and a true / false value indicating whether the attribute is true or false.
- a boolean value can have one of three types of values: true, false, or neither false. For example, the true value is “1”, the false value is “ ⁇ 1”, and the value when neither true nor false is “0”.
- an object having “1” as a truth value has an attribute represented by the coordinate axis.
- an object having “ ⁇ 1” does not have the attribute, and an object having “0” indicates that it is not characterized by the attribute.
- the true / false value can also be set to “0” when it is not known whether it is true or false, that is, when it is indefinite. It is natural that a pair of one attribute identifier for identifying an attribute and one true / false value becomes one attribute. Instead of the values “1”, “0”, “ ⁇ 1”, a truth value may be used as an identifier representing the truth value.
- any of the three types of identifiers that are not “true”, “false”, and “true or false” can be set as a true / false value.
- “true” may be “20005”
- “false” may be “20006”
- “not true or false” may be “20007”.
- Attribute can have no Boolean value. If an object has an attribute, the truth value of that attribute can be true. On the other hand, when a thing does not have an attribute, there is a method in which the true / false value of the attribute is set to false. There is also a method in which a true / false value is false or indefinite when an object does not have an attribute. Indefinite means that it is not determined whether it is true or false. In this method, since there is no true / false value of the attribute, the logical operation process is simplified. In the world of mathematics, the discovery of the number 0 is a big event. Clearly distinguishing the state that an object does not have an attribute by having a false value of false may be a sophisticated way of holding information. Therefore, when it is limited to the fourth information processing, it may be more natural that the attribute does not have a true / false value.
- mountains, rock salt, and mountains made of rock salt can be described as follows.
- the coordinate axis 1 of the mountain is 1
- For rock salt, coordinate axis 2 is 1, coordinate axis 3 is -1.
- a mountain made of rock salt, coordinate axis 1 is 1, coordinate axis 2 is 1, coordinate axis 3 is -1.
- the same content can be described using symbols as follows.
- Mountain made of rock salt (1: 1, 2: 1, 3: -1)
- the left parenthesis is a thing, the whole parenthesis represents a set of attributes that the word has, and the parentheses represent individual attributes that the thing has.
- a number on the left side of “:” in parentheses is an attribute identifier representing an attribute, and a right side of “:” is a true / false value of the attribute. If an object is not characterized by a certain attribute, the attribute identifier and truth value description for that coordinate axis are omitted. Having no attribute means that the attribute does not affect the meaning of the object. Or it shows that there is no information about the relation between the object and its attribute.
- the attribute may be described only when the object has an attribute, that is, only when the true / false value of the attribute of the object is true. In this case, it has no truth value. This may be more natural in the fourth information processing. This is closer to the above hypothesis.
- An attribute that does not contain a true / false value is described only when the true / false value of the attribute of the object is true.
- the attribute identifier representing the attribute can be composed of a main identifier and a sub-identifier.
- the primary identifier represents the basic property of the attribute
- the secondary identifier represents a more detailed property of the attribute such as degree and appearance.
- the sub-identifier corresponds to an adverb.
- the above-mentioned hypothesis main attribute identifier corresponds to the attribute main identifier
- the above-mentioned hypothesis sub-attribute identifier corresponds to the attribute sub-identifier.
- the attribute is composed only of the main attribute.
- the main attribute and the sub attribute are attributes and have the same data structure.
- the attribute of height can have more detailed attributes such as tall, slightly tall, and short.
- a main identifier that is an identifier of a main attribute can be associated with height that is a basic attribute.
- an attribute that modifies the basic attribute of high height and low height can be associated with a sub-identifier that is a sub-attribute identifier.
- feature data generated from an image in which a person's height is measured with a height meter is directly connected to the main attribute and associated.
- Feature data generated from a small person's video is directly linked to the sub-attribute and associated.
- the feature data directly associated with the sub-attribute includes not only numerical feature data but also feature data generated from information obtained from sensory organs and sensors called the five senses such as sound and video.
- the sub-identifier that is an identifier of the sub-attribute modifies the main identifier that is the identifier of the main attribute like an adverb.
- the main identifier as a key
- feature data directly corresponding to the main identifier of the attribute can be acquired.
- feature data directly corresponding to the sub-identifier of the attribute can be acquired.
- the sub-identifier corresponding to “Very salty” is set to 9
- the sub-identifier corresponding to “feeling a little salty” is set to 1.
- a plurality of sub-identifiers may be provided in stages between “feel a little salty” and “very salty”. This is expressed as follows. Since the sub-identifier is also an identifier, it may be an identifier indicating data indicating the feature of “very salty”, for example, feature data having a value of 9. In this example, the data representing the feature of “very salty” and its sub-identifier are represented by the same character as intuitively understood. (1) Mountain (1: 1) (2) Rock salt (2 * 9: 1, 3: -1) (3) Mountain made of rock salt (1: 1, 2 * 9: 1, 3: -1)
- a mountain is represented by A
- B a rock salt
- C a mountain made of rock salt
- “A” and “B” are not “words” indicating the meaning of the object, but are “object identifiers” that do not represent the meaning of the object but simply identify the object.
- the thing represents that it consists of one “object identifier” and one attribute set.
- an identifier of 30003 may be used. Since a set can have zero or more attributes as elements, the number of attribute identifiers representing the attributes of a “thing identifier” representing a thing is not limited to one, but two pieces, either zero or one That's fine.
- objects can be expressed as points, lines, planes, solids, etc. in a space with attributes as coordinate axes.
- the degree of overlap or proximity of two objects represents the magnitude of the relationship between the two objects.
- objects and attributes associated with brain neurons can be expressed well.
- an object or attribute By representing an object or attribute with an identifier for identifying the object or attribute, it is possible to express a connection between the object or attribute associated with a neuron in the brain.
- searching for an object or attribute using an identifier for identifying the object or attribute as a key the connected object or attribute is acquired.
- an object identifier connected to the attribute can be acquired.
- a table that associates attribute identifiers with feature data is created in advance for connected attributes and feature data.
- this table is searched using the attribute identifier as a key, feature data connected to the attribute can be acquired.
- feature data connected to the attribute can be acquired.
- an attribute identifier connected to the feature data can be acquired. In this way, the connection of things, attributes, and feature data can be imitated.
- the neurons corresponding to the object or attribute are not connected to each other by dendrites or the like, but the object or attribute is represented only by an identifier that does not express its meaning, and the identifier is used as a key.
- identifiers In conventional information systems, people are associated with identifiers called employee numbers. However, in the prior art, a person can have a plurality of identifiers of different formats such as an annuity number in addition to the identifier of an employee number.
- An automobile has an identifier called a license plate number or a manufacturing number.
- the employee number, pension number, and car number are different types of information. Even the same person can have different types of identifiers. Different objects, such as people and cars, have different types of identifiers.
- the true / false value may be an identifier representing the true / false value.
- the true / false value is either an attribute representing true or an attribute representing false.
- An attribute representing true is connected to feature data representing true, and an attribute representing false is associated with feature data representing false.
- All identifiers including the primary attribute identifier that is the primary attribute identifier, the secondary attribute identifier that is the secondary attribute identifier, and the true / false identifier can be limited to the same format as the entity identifier.
- features are extracted from the received information.
- the extracted features are stored as feature data.
- the retained feature data is associated with the object identifier and the attribute identifier.
- identification data connected to an object or attribute is acquired, a word is generated from the acquired identification data, and the generated word is output.
- Identification data generated from the words “flower” in Japanese and “flower” in English is connected to one attribute.
- the feature data generated from the image of the actually blooming flower is associated with another attribute identifier.
- B is a word or identifier that represents an object and has a set of attributes.
- the argument (2 * 9: 1) is a set of attributes including one attribute.
- a set of attributes of the object is an operation target. In this case, it happens to be a set of one attribute, but there may be zero or a plurality of attributes.
- the program “include ((2 * 9: 1))” returns a true value as a return value if the attribute set of B has the same attribute as the argument. If there is the same attribute identifier but the truth value is different, a false value is returned as a return value.
- NULL may be used as a value indicating that it does not exist. Further, when there is no same attribute identifier, a false value may be returned as a return value.
- a program called include performs the following process to generate a true or false value or a NULL value.
- a thing having a set of attributes to be compared, BverySalt, that is, a set of attributes that are arguments of the program is called a set of attributes to be compared.
- BverySalt that is, a set of attributes that are arguments of the program
- a true value is returned as a return value.
- a false value is returned as a return value.
- NULL is returned as a return value in the sense that there is an attribute that cannot be compared. If the attribute set to be compared has no attribute in the set of attributes to be compared, false may be returned. Whether NULL is returned or false is returned depends on the design policy of the program.
- the return values and the conditions listed here are examples. A combination of other conditions and return values may be used.
- condition with false and NULL as return values is shown below. It is merely an example and other conditions may be used. Returns a false value as a return value if the primary identifier of the attribute of the set of attributes to be compared is in the set of attributes to be compared and if there are fewer than a predetermined number of matching true / false values . If there is no primary attribute identifier in the set of attributes to be compared, the NULL value is returned as a return value in the sense that comparison is not possible.
- the above-mentioned conditions are examples, and other conditions may be used as long as they are appropriate.
- Attribute sub-identifiers should indicate value, value range, degree, status, etc. as supplementary information.
- a sub-identifier of an attribute that is, the sub-attribute is connected to a value, a range of values, a degree, and a situation as feature data. It is also possible to specify a main identifier, and use the determination whether the main identifier is the same and the sub-identifier is the same. If values and ranges, which are feature data corresponding to the sub-identifiers, are stored in advance, and information on the values and ranges is acquired using the sub-identifiers as keys, the values can be directly compared.
- the main attribute of height can be connected to a sub-attribute connected to feature data of a specific value, for example, 190 cm.
- a specific value for example, 190 cm.
- attributes that are not clear such as the attribute of being tall. Considering Japanese men as a premise, almost all people judge that they are 185cm tall. 60% of people may determine that they are 180cm tall. As in this example, attributes often do not have clear boundaries.
- Numerals, audio data, image data, video data, etc. are also included in the actual data.
- the feature data generated from the actual data is connected to the same attribute together with the fuzzy function.
- Actual data is not limited to the aforementioned data.
- the ratio of width to length can be obtained from face image data, and the degree of the round face can be generated using the held membership function with the ratio as an argument.
- FIG. 14 is an example of a fuzzy set in one embodiment.
- the membership function is generated by generating values from 0 to 1 representing the degree indicated by the vertical axis in FIG. 14 when data such as a specific value which is the value on the horizontal axis in FIG. 14 is given. Returns a value representing the degree as a return value.
- a method of generating unclear attributes there is a method of connecting various feature data and identification data to one attribute. For example, in speech recognition, a certain word is pronounced by many people, and identification data is generated from the pronunciation. The plurality of generated identification data are connected to one attribute. This attribute represents the phonetic representation of the word. When the same identification data as any of the identification data connected to the attribute is generated from the input voice, the input voice is recognized as the attribute.
- This method can also be used for feature data, and can be applied to all kinds of information such as images in addition to audio.
- This common degree is defined as a common degree. For example, with respect to a set of two attributes, it is checked whether there is an attribute having the same attribute identifier and true / false value. Thus, if a program that returns the number of sets of matching attributes as a common degree is created, the common degree can be obtained. When an attribute has a weight as attached information, the commonness may be obtained by adding the weights of matching attributes. Adding the weights is also included in the counting. The method for calculating the commonality is an example. Other methods may be used.
- the degree of disagreement is defined as the non-commonness. For example, it is checked whether there is an attribute having the same attribute identifier and different true / false values with respect to a set of two attributes. To. Thus, if a program that returns the number of sets of corresponding attributes as the non-commonness is created, the non-commonness can be obtained. When an attribute has a weight as attached information, the degree of dissimilarity may be obtained by adding the weights of the corresponding attribute.
- the non-commonness calculation method is an example. Other methods may be used.
- ⁇ Commonness and non-commonness can be used as a measure to indicate how similar or how different two attribute sets are.
- a list of matching attributes can be returned as a return value.
- a list of non-matching attributes it is also possible to return a list of non-matching attributes as a return value.
- the degree of commonality is large, it means that there are many common attributes. When the non-commonness is larger than 0, it means that the conflicting attribute may be included. When reading a book or listening to a story, you may feel "Oh, something strange”. This feeling can be associated when the non-commonness is greater than zero. If the degree of commonality is large, but the degree of commonality is not 0, there may be some inconsistency.
- AND operation newly generates a set of attributes whose elements are attributes common to the two sets of attributes.
- a program that performs an AND operation collects attributes having the same attribute identifier and the same true / false value for two attribute sets, and generates a new attribute set having the collected attributes as elements.
- the AND operation is also a product set operation.
- c is a set of attributes not including attributes.
- the symbol ⁇ represents a set of attributes having no attribute.
- the OR operation newly generates a set of attributes including all the attributes included in at least one of the two sets of attributes.
- a program that performs an OR operation generates a new attribute set whose elements are all the attributes included in the two attribute sets.
- the OR operation is also a union operation.
- the lowercase alphabet is a set of attributes
- the uppercase alphabet is a thing
- the logical calculation formula does not change even if the number of attributes included in the attribute set is large or small. Usually, things have very many attributes. If this technology is used, a logical operation expression can be described very simply as compared with the conventional technology.
- an object and an attribute of the object are not expressed by a word but by an identifier that is information having no meaning of the object or the attribute itself.
- a set of attributes whose elements are attributes represented by identifiers, which are information that does not represent the meaning of the attributes themselves, is the target of the logical operation.
- Each attribute that is an element of a set of attributes is associated with one coordinate axis.
- a coordinate axis identifier for identifying a coordinate axis corresponding to each attribute is used as an attribute identifier, and the attribute is represented by a combination of the attribute identifier and a true / false value. If the attribute is true, the attribute is represented by a combination of an attribute identifier and a positive value. On the other hand, if the attribute is not true, the attribute is represented by a set of an attribute identifier and a negative value. This interpretation may be easy to understand.
- the purpose of associating attributes with coordinate axes is to aid understanding of the present technology. It is not essential to associate attributes with coordinate axes.
- the attribute may be an attribute identifier composed of a main identifier and a sub-identifier.
- the primary attribute identifier corresponds to the primary identifier
- the secondary attribute identifier corresponds to the secondary identifier.
- a coordinate axis identifier for identifying a coordinate axis corresponding to each attribute is a main identifier, a value indicating a more detailed feature of the attribute, a range, a degree, etc. are sub-identifiers, and an identifier composed of a main identifier and a sub-identifier It is good also as an attribute which consists of a truth value.
- a special value indicating that there is no sub-identifier for example, a special value such as NULL may be used as the sub-identifier. Alternatively, it may simply not have a sub-identifier.
- the method of expressing the degree of attribute using the sub-identifier has been shown.
- the criteria for determining the attribute of being tall is different depending on the race. Therefore, it is beneficial to be able to acquire actual information such as height from the attributes. It is only necessary to store actual height numerical information as feature data, generate an attribute called height, and associate the feature data of height with the sub-attribute.
- the feature data generated from the whole body image may be connected to the attribute of height. Based on the feature data connected to the attribute of height, it is possible to determine whether the person is tall and change the meaning of the attribute. In this way, actual information can be acquired from the attribute identifier. Using the attribute identifier as a search key, actual information, that is, feature data can be acquired.
- Attribute can be generated from all kinds of information such as images, videos, sounds, touch, temperature and pressure.
- attributes such as face length and round face can be generated from a face image. It is assumed that a face portion is extracted from image data, and an aspect ratio calculation function for calculating a ratio between the width and length of the face from the extracted face image data is provided.
- face image data is given
- the aspect ratio of the face is calculated from the given face image data using the aspect ratio calculation function.
- the attribute closest to the calculated aspect ratio of the face is selected, and the truth value of the selected attribute is made true.
- identification data can be generated from a word and held, and the identification data generated from the word can be associated with an attribute.
- the identification data generated from words is merely used to select an object or attribute, and is not used to represent the meaning or concept of the object or attribute.
- an image diagnosis system can generate an attribute in which feature data generated from information other than words such as images, sound, touch, temperature, and pressure is connected, and perform a logical operation on a set of attributes including the generated attribute. It can be used for equipment monitoring systems.
- a thing may inherit the set of attributes of another thing.
- An object may also contain a set of attributes of another object.
- the OR operation is associated with inheriting or including the attribute of another object.
- a collection of things with common attributes is called a knowledge collection. Specify one or more attributes and select the one with the specified attribute identifier.
- a collection of selected objects is a collection of knowledge. Normally, an object having the attribute identifiers of all the specified attributes is selected, but an object having a part of the specified attributes may be selected.
- it may be a collection of knowledge consisting of only one thing. In this case, it can be interpreted that one object is identified and recognized by the attribute used for the search.
- the thing becomes an abstract object that means a collection of things.
- the abstract object can be identified by the identification data.
- one thing and another thing may have a relationship. Connect these related objects.
- This collection of connected objects is a knowledge network. Expressing the relationship between things by connecting things. There are two types of relationships: one is connected in both directions and the other is connected in one direction. When connected in both directions, the two connected objects cannot be distinguished from each other, so that the relationship is equal. When connected in one direction, two connected objects can be distinguished. For example, the parent-child relationship is represented by a one-way connection, and the parent thing is connected in one direction only to the child thing. In this way, it is possible to distinguish a parent and a child.
- a relationship represented by a one-way connection can be associated with an object being included in another object, such as an inclusion relationship.
- adding a connection between things, changing a connection method, and deleting a connection are associated with learning knowledge.
- Adding new objects to the knowledge network or deleting existing ones can also be associated with knowledge learning.
- Creating a new knowledge network is also associated with learning.
- the existence and connection of things in the knowledge network is also knowledge. Things are also knowledge.
- FIG. 15 shows the inclusion relationship of objects.
- a circle shown in FIG. 15 is a node 60.
- the node 60 holds an object identifier and represents an object.
- the node 60 is only the thing represented by the thing identifier.
- the attribute 61 and the set of attributes 61, feature data, and identification data are not the node 60.
- the attribute 61 and the set of attributes 61 are attached information of an object that is the node 60.
- Objects include not only concrete objects such as stones and dogs, but also shining objects and abstract objects such as WEB pages and concepts.
- a node 60 having no attribute 61 is also possible. If there is an object that is not known, the object does not have the attribute 61.
- the node 60 has zero or more attributes 61.
- An entire attribute of one object can be regarded as a set of attributes 61.
- the set of attributes 61 corresponds to the list of objects and attributes shown in FIG.
- An arrow 62 represents the association of the node 60. That is, the arrow 62 represents the association of objects.
- An arrow 63 represents the connection between the node 60 and the attribute 61.
- the second method is as follows.
- the thing of a subordinate concept does not have the attribute showing a superordinate concept.
- the attributes representing the plurality of concepts can be arranged in the same order by connecting them in order from the attribute representing the higher concept to the attribute representing the lower concept.
- An attribute after a certain attribute is selected from the hierarchical information of the attribute, and a collection of knowledge is generated by collecting objects having at least one selected attribute. If a collection of knowledge is generated by selecting all attributes after a certain attribute from the hierarchical information of the attribute, a collection of knowledge including all subordinate concepts can be generated.
- you select a range of attributes you can generate a collection of knowledge with that range of attributes.
- objects can be connected in a collection of knowledge.
- a connected object is called a knowledge network.
- things that do not have common attributes that is, things that are not included in the knowledge collection, can be connected to the knowledge network.
- knowledge networks In addition to a tree structure in which one or more nodes are connected to one end node, there are knowledge networks of various connection methods. There are connections with direction and connections without direction. If there is a direction, it can indicate that there is a difference between the two connected objects. Connections without direction can be viewed as being connected to both rather than one direction. Directional connections are represented by arrows. Connections without direction are represented by lines.
- An arrow has a starting point and an arriving point, and the one with the arrow is the arriving point, and the one that is not the arriving point without the arrow is the starting point.
- the starting point may be called the start point of the arrow, and the arrival point may be called the end point of the arrow.
- a knowledge network with a network structure without a root node as a starting point such as a tree structure is also possible. If things are connected, it is a knowledge network. Any connection method is acceptable. There can be only one thing or a zero knowledge network.
- a connection by a dendrite or the like connecting nerve cells corresponds to an arrow.
- Objects can belong to multiple knowledge collections.
- An object can belong to a plurality of knowledge networks.
- humans have an attribute called blood relationship. It also has attributes like a school of origin. If these multiple attributes are expressed by a single knowledge network, the knowledge network becomes very complex.
- a knowledge network representing a blood relationship and a knowledge network representing a home university are respectively generated so that a human node belongs to a plurality of knowledge networks such as a knowledge network representing a blood relationship and a knowledge network representing a school of origin.
- nodes that are objects are connected by a blood relationship.
- nodes that are grades are connected.
- the structure of each knowledge network becomes very simple if the connection between things is expressed by dividing it into several knowledge networks.
- things belong to multiple knowledge networks.
- knowledge networks things are directly connected by some relationship.
- a knowledge network can express clearly systematized knowledge by connecting things having common attributes by only one predetermined relationship.
- systematized knowledge about the object can be expressed and maintained.
- Knowledge networks are connected only by things.
- a knowledge collection is created by collecting objects having all the specified attributes, and if it is determined whether there is a given object in the generated knowledge collection, then the given object is the knowledge of the knowledge. You can determine if you have all the attributes that generated the collection.
- a knowledge collection is a collection having an object (object identifier) as an element. Therefore, logical operations such as an AND operation, an OR operation, and a NOT operation are possible using an abstract object representing a collection of knowledge as an operation target.
- an arrow connecting nodes in a certain knowledge network represents an inclusion relationship.
- the starting point node includes the arriving point node. If node B can move to node A more than 0 times in the opposite direction of the arrow and reach node A, B is said to be A. When the number of movements is 0, node A and node B are the same node. Again, nodes represent things.
- node B may be node D.
- Node D there may be some relationship between Node B and Node D. The degree of possibility depends on how well the knowledge network is generated. When the total number of movements is 0, node B, node A, and node D are the same node.
- the node D can be connected to the node B with one arrow.
- a new knowledge network including node D and node B may be created, and node D and node B may be directly connected.
- a one-way connection knowledge network was used as an example, but a two-way connection knowledge network is also possible. However, in the case of two-way connection, the difference between connected objects is not clear.
- the relational database 14 can efficiently perform set operations such as extracting data that satisfies certain conditions from tabular data. Therefore, in this embodiment, an example using the relational database 14 will be described. You may implement without using the relational database 14.
- FIG. In the technology of this application, instead of knowing the connected object or attribute by connecting the object and attribute and sending a signal to the connected object or attribute, the object, attribute, feature data, and identification data are represented by an identifier. And search. As a result, connected objects, attributes, feature data, and identification data are acquired.
- the object identifier itself is an object
- the attribute identifier itself is an attribute
- the primary identifier of the attribute itself is the primary attribute
- the secondary identifier of the attribute itself is the secondary attribute.
- the feature data is represented by a pointer that is an identifier.
- the feature data itself is in data 10.
- the identification data is represented by a pointer that is an identifier.
- the identification data itself is in data 10.
- 17A, 17B, 18 and 19 schematically show the tables in the data 10 created by the relational database 14.
- FIG. 17A is a list of data that is the basis of the attribute.
- FIG. 17B is a correspondence table between the names of the objects and the object identifiers.
- the data list that is the basis of the attribute includes a word that is the name of the object represented by characters, a word that represents the attribute of the object, and a truth value of the attribute.
- the actual data is data that is a source of the truth value. If there is actual data, a true / false value may be generated from the actual data.
- a Boolean value is not required. By applying the rule that a boolean value is true only when the attribute is listed in the table, the boolean value can be omitted.
- the correspondence table between the object name and the object identifier associates a word that is the name of the object with the object identifier.
- FIG. 18 is a correspondence table of attributes and attribute identifiers associated with words.
- This table corresponds to the first hierarchy in FIGS. 2A and 2B.
- the attribute identifiers in this table are associated with feature data stored in advance that is not shown.
- a main identifier that is a main attribute is associated with a pointer that is an identifier of feature data of the main attribute.
- a sub-identifier that is a sub-attribute is associated with a pointer that is an identifier of feature data of the sub-attribute.
- the word associated with the attribute is associated with the attribute identifier.
- the identifier of feature data that is a pointer to the feature data is also associated.
- Attribute attribute and feature data outside the figure are connected, meaning or concept is generated.
- the number of feature data pointers to be associated may be limited to one at a maximum, and as many table data rows as the number of feature data to be associated may be created, or all feature data pointers may be included in one row.
- the correspondence table between attributes and attribute identifiers is a table for associating the attribute words in the data list that is the basis of the attributes in FIG. 17A with the attribute identifiers.
- the corresponding attribute identifier is acquired from the attribute-attribute identifier correspondence table using the word associated with the attribute in the data list as the attribute source in FIG. 17A as a key.
- the attribute identifier to be acquired may consist of a main identifier and a sub-identifier. Sometimes it consists only of the primary identifier. If a possible combination of a main identifier and a sub-identifier is predetermined, an unnatural combination can be detected using the combination. Even if possible combinations of primary identifiers and secondary identifiers are created in advance, possible combinations of primary identifiers and secondary identifiers can be added or deleted by learning.
- the attribute of the data list that is the basis of the attribute in FIG. 17A may be an attribute composed of a main attribute and a sub attribute.
- the correspondence table of attributes and attribute identifiers in FIG. 18 may have a condition that the true / false value of the attribute is true. If the correspondence table of attributes and attribute identifiers has a condition that the true / false value of the attribute is true and has actual data that is the attribute's source, the true / false value of the attribute is true.
- the true / false value can be determined by judging whether or not the condition is satisfied.
- the condition may be a determination function such as a membership function of a fuzzy set. In this case, the data on which the attribute is based may not have a true / false value.
- Feature data is generated from the actual data.
- 17A is a list of data that is the basis of attributes, a correspondence table between the names of the objects and the identifiers in FIG. From the identifier, the attribute identifier corresponding to the attribute, and the true / false value, as shown in FIG. 19, a list of the thing and the attribute that associates the object identifier, the attribute identifier of the thing and the true / false value, Generate. Boolean values can be omitted by applying the rule that true / false values are true if there is an entity identifier and the attribute identifier of the attribute that the entity has, and false otherwise. . Further, feature data generated from actual attribute data may or may not be included. Note that having an identifier of feature data also describes having feature data. When the attribute consists of one main attribute and one sub attribute, the main attribute and the sub attribute are distinguished by the column position of the table.
- feature data outside the figure is associated with an object identifier and an attribute identifier. Then, by searching using the identifier of the feature data as a key, an object identifier and an attribute identifier associated with the feature data can be acquired.
- Objects and attributes connected to feature data have feature data identifiers that are pointers to the feature data.
- FIG. 20A is a table in which objects, main attributes, and sub-attributes are associated with each other.
- This table is an example showing a set of one thing, one main attribute, and one sub attribute connected so as to form a triangle having the object, the main attribute, and the sub attribute as vertices. If this table is searched by specifying an object and a main attribute, sub-attributes that are remaining elements forming a triangle can be acquired.
- the main attribute and the sub attribute are associated with each other, and an appropriate combination of the main attribute and the sub attribute is held.
- the number of sub-attributes associated with one main attribute is not limited to one.
- the number of sub-attributes associated with a certain main attribute is not limited to one, and there may be a plurality.
- Choose an appropriate combination of main attribute and sub-attribute from this combination associate the object with the main attribute and sub-attribute, and list the associated object, main attribute and sub-attribute pair with the object and attribute. Can be added to.
- the same feature data may be directly connected to and associated with the main attribute and the sub attribute constituting one attribute.
- the number of feature data associated with the main attribute and the sub-attribute constituting one attribute is not limited to one, and there may be a plurality of feature data.
- Both the main attribute and the sub attribute are attributes, and both are one or more, and feature data that is not limited to one is connected.
- object, main attributes, and sub-attributes are connected in a triangle shape, if you specify two of the three elements of the objects, main attributes, and sub-attributes that make up the triangle, the remaining one element is determined.
- an attribute connected to a certain object is composed of a main attribute and a sub-attribute, it is possible to search and acquire an attribute connected to both the object and the main attribute.
- This attribute is a sub attribute.
- the main attribute is connected to feature data representing a rough meaning or concept.
- the sub-attribute is connected to feature data representing a more detailed meaning or concept. If you want to know a rough meaning or concept, you need to acquire feature data that leads to the main attribute. If you want to know more detailed meaning or concept, you need to acquire the feature data that leads to the sub-attribute.
- This operation of selecting a common attribute from two attribute sets and generating a new attribute set having the selected attribute as an element is called an AND operation.
- An operation of selecting an attribute included in at least one of the two attribute sets and newly generating a set having the selected attribute as an element is called an OR operation.
- For an attribute that is an element of a set of attributes if the truth value is true, the truth value is set to false, and if the truth value is false, an attribute with the truth value set to true is generated.
- An operation for newly generating a set of attributes having attributes as elements is called a NOT operation.
- An operation of selecting an attribute included in only one of the two attribute sets and generating a new attribute set having the selected attribute as an element is called an XOR operation.
- FIG. 21 is a flowchart of an AND operation according to an embodiment. The processing procedure of the AND operation will be described with reference to FIG.
- an attribute set A and an attribute set B to be ANDed are designated (S21).
- a set of zero or more attribute identifiers associated with the object identifier is specified.
- an attribute having the same attribute identifier is selected from the attributes belonging to the attribute set A and the attributes belonging to the attribute set B (S22).
- an attribute having the same attribute identifier and the same associated true / false value is selected, and a set of attributes having the selected attribute as an element is newly generated (S23). Then, a newly generated set of attributes is set as a result of the AND operation.
- the attribute does not have a true / false value, an attribute having the same attribute identifier is selected, and a set of attributes having the selected attribute as an element is newly generated. It is described that two matching processes are combined to select matching attributes. Then, the AND operation selects a matching attribute for the attribute of the attribute set A and the attribute of the attribute set B, and newly generates an attribute set (common part) having the selected attribute as an element. It is.
- FIG. 22 is a flowchart of an OR operation in one embodiment. The procedure of OR operation will be described with reference to FIG.
- an attribute set A and an attribute set B to be ORed are designated (S31).
- an attribute having the same attribute identifier and having both an affirmative value and a negative value is selected from the attributes of the attribute set A and the attribute set B (S32). If the attribute does not have a true / false value, an attribute having the same attribute identifier is selected. It is described that two matching processes are combined to select matching attributes. This process coincides with the result of the AND operation described above.
- an attribute that is not included in the attribute set B among the attributes in the attribute set A is selected. That is, an attribute identifier included only in the attribute set A is selected.
- an attribute that is not included in the attribute set A among the attributes of the attribute set B is selected. That is, an attribute identifier included only in the attribute set B is selected (S33).
- a new set of attributes whose elements are all the attributes selected in S32 and S33 are generated as a result of the OR operation (S34).
- the attribute set obtained by the above processing is a new attribute set (union set) in which attribute identifiers included in at least one of attribute set A and attribute set B are collected. It becomes.
- the attribute set obtained by the above processing is, for example, from the sum of attribute set A and attribute set B to either attribute set A or attribute set B. It can also be obtained by removing attributes that are included and associated with different truth values.
- FIG. 23 is a flowchart of a NOT operation according to an embodiment. The NOT calculation processing procedure will be described with reference to FIG.
- a set of attributes to be operated is designated (S41).
- the value of the associated truth value is inverted for all attributes included in the specified attribute set. That is, when the true / false value is a true value, a new attribute is generated, and when the true / false value is a false value, a new attribute is generated. Then, the generated attributes are collected to generate a new attribute set (S42).
- FIG. 24 is a flowchart of NAND operation in one embodiment. A processing procedure of NAND operation will be described with reference to FIG.
- an attribute set A and an attribute set B to be subjected to NAND operation are designated (S51).
- an attribute having the same attribute identifier and the same associated truth value is selected to generate a new attribute set ( S52). This corresponds to the result of the above AND operation.
- FIG. 25 is a flowchart of the XOR operation in one embodiment. The processing procedure of the XOR operation will be described with reference to FIG.
- an attribute set A and an attribute set B to be subjected to the XOR operation are designated (S61).
- an attribute that is not included in the attribute set B is selected among the attributes of the attribute set A. That is, an attribute identifier included only in the attribute set A is selected.
- an attribute that is not included in the attribute set A among the attributes of the attribute set B is selected. That is, an attribute identifier included only in the attribute set B is selected (S62).
- This process corresponds to the process of S33 shown in FIG.
- a set of attributes having all the attributes selected in S62 as elements is newly generated as a result of the XOR operation (S63).
- a new attribute set is generated by collecting attribute identifiers included only in one of the attribute set A and the attribute set B.
- the attribute set obtained by the above processing is, for example, from the union of the attribute set A and the attribute set B (the result of the OR operation) to the common part of the attribute set A and the attribute set B (AND operation). It is also possible to obtain the result by removing the result.
- FIG. 26 is a flowchart of a process for calculating the commonality according to an embodiment. With reference to FIG. 26, a processing procedure for calculating the commonality will be described.
- an attribute set A and an attribute set B to be operated are designated (S71).
- an attribute having the same attribute identifier and the same associated true / false value is selected (S72). If the attribute does not have a boolean value, select an attribute with the same attribute identifier. It is described that two matching processes are combined to select matching attributes. Even if the sub-identifiers are different, the attributes may be the same if the main identifiers are the same.
- the result of counting the number of selected attributes is output as the degree of commonality (S73).
- the condition for selecting the matching attribute is an example, and other conditions may be used.
- the weight values may be summed instead of counting the number of pairs.
- FIG. 27 is a flowchart of a process for calculating non-commonality in an embodiment. With reference to FIG. 27, a processing procedure for calculating the non-commonality will be described.
- an attribute set A and an attribute set B to be operated are designated (S81).
- an attribute having the same attribute identifier and having a different associated true / false value is selected from the attributes of attribute set A and attribute set B (S82). If the attribute does not have a true / false value, an attribute that exists only in one attribute set is selected.
- the two selection processes are described as selecting mismatched attributes.
- the result of counting the number of selected attributes is output as the non-commonness (83).
- the method of selecting the non-matching attribute is an example, and other selection conditions may be used.
- the weight values may be summed instead of counting the number of pairs.
- processing for each attribute in the attribute set is automatically performed by the arithmetic unit 1070 held in advance.
- the user of this logical operation system does not individually instruct processing for each attribute.
- the operation is automatically performed on the attribute of the object or the attribute of the attribute set.
- Objects are stored in the database, for example, in a data structure called a list of objects and attributes shown in FIG.
- One or more attributes are specified, and an object having all the specified attributes is selected from the attribute list. Then, a collection of objects composed of objects having all the specified attributes is generated. This is called a knowledge collection.
- a knowledge collection When all the attributes used to generate a knowledge collection are connected to a single object, the object becomes an abstract object that means a knowledge collection.
- an attribute that directly connects identification data that identifies an abstract object that represents a collection of knowledge or an attribute that directly connects feature data that characterizes an abstract object that represents a knowledge collection represents an abstract object that represents that knowledge collection. Connect. Then, it becomes easy to distinguish an abstract object representing the knowledge collection from an object included in the knowledge collection.
- the knowledge network represents systematized knowledge. You can create a knowledge network that is linked by age in the knowledge gathering of a family, or you can create a knowledge network by a parent-child relationship. A plurality of knowledge networks can be created in one knowledge collection.
- An object in a knowledge network in one knowledge collection may belong to another knowledge network in another knowledge collection.
- a single person may belong to a knowledge network called a family and at the same time belong to a knowledge network called a school.
- 28A to 28C show an embodiment of a data structure for realizing the knowledge network of FIG.
- FIG. 28A is a correspondence table 86 between the object name 87 and the object identifier 88, and corresponds to the correspondence table between the object name and the object identifier in FIG. 17B.
- the correspondence table 86 associates a word, which is a character string describing the name of an object in words, with an object identifier 88.
- FIG. 28B is a correspondence table 101 between knowledge networks and attributes, and associates the attribute identifier 103 that characterizes the knowledge network with the knowledge network identifier 102.
- a weight may be directly associated with each row of the correspondence table 101, that is, with each knowledge network identifier 102 and attribute identifier 103 as attached information. This weight represents the strength of the relationship between the knowledge network and the attribute.
- the node 80 which is an object in the knowledge network has only one object identifier 81. Further, the node 80 that is a thing can have knowledge network information 84. For each knowledge network to which a node that is a thing belongs, one piece of knowledge network information is held in the node that is the thing.
- the number of the knowledge network information 84 possessed by the node 80 as one object may be zero or one, and is not limited to one and may be plural.
- the knowledge network information 84 includes zero or more forward pointers 82, zero or more backward pointers 83, and one knowledge network identifier 85. When the forward pointer 82 and the backward pointer 83 are both zero, the node 80 which is a thing becomes one knowledge network.
- node A which is a thing
- node B which is a thing
- node A holds the object identifier of node B in forward pointer 82
- node B holds the object identifier of node A in backward pointer 83. To do. If it does in this way, the direction of a connection can be expressed, and it can move easily in the forward direction which is the direction of connection, and the reverse direction which is the reverse direction of connection.
- the node A holds the node B object identifier in the forward pointer 82, and the node B stores the node A object in the forward pointer 82. Holds the identifier.
- the start point of a pointer with direction may be distinguished from the upper side and the end point may be referred to as lower, or the start point of the direction pointer may be referred to as lower and the end point as upper. Also, when connected by a bidirectional pointer, there is no distinction between upper and lower, so it may be called parallel.
- nodes A and C which are two connected objects, each have a forward pointer 82 connected to the other, they are connected in both directions. If one of the two connected nodes A and C holds the other object identifier in the forward pointer 82 and the other holds the one object identifier in the backward pointer 83, they are connected in one direction. .
- the knowledge network information 84 of the node 80 may further have a weight 100, and the weight 100 may represent the strength of the relationship between the knowledge network and the object.
- the weight 100 may not be retained.
- the node 80 that is a thing holds the knowledge network identifier 85 and the weight 100 together, thereby expressing the strength of the relation between the thing and the knowledge network.
- this weight 100 can be used as a criterion. Select a knowledge network having a larger weight 100 value, select a knowledge network having a smaller weight 100 value, or select a knowledge network in a range in which the weight 100 value is specified.
- the weight 100 can be used as a selection criterion.
- each of the forward pointer 82 or the backward pointer 83 of one knowledge network information 84 has a weight and represents the degree of strength of connection with the node indicated by the pointer. May be. Thereby, the weight can be used as a reference when one pointer is selected from several pointers.
- Knowledge network identifier 85 is an identifier for identifying a knowledge network.
- a knowledge network in a knowledge collection generated by specifying a plurality of attributes is characterized by the plurality of attributes. There can of course be a knowledge network characterized by one attribute.
- the knowledge network identifier 85 When the knowledge network identifier 85 is connected to all the attributes that characterize the knowledge network, the knowledge network identifier 85 has an attribute that characterizes the knowledge network and becomes an abstract object representing the knowledge network. Attributes other than the attribute used for selection may be connected to the knowledge network identifier.
- a knowledge network created regardless of the knowledge collection may identify the knowledge network with an attribute representing the characteristics of the knowledge network.
- an attribute characterizing each knowledge network is added to distinguish them. In this way, it becomes easy to select one knowledge network from a plurality of knowledge networks in one knowledge collection.
- FIG. 29 shows an embodiment of the data structure of the node.
- the correspondence table between objects and knowledge networks shown in the upper part of FIG. 29 associates knowledge network identifiers with object identifiers.
- the knowledge network identifier to which the object identified by the object identifier belongs is known.
- the knowledge network identifier, the object identifier, and the weight may be associated with each other. Of course, this weight may be omitted. This weight represents the degree of relevance between the knowledge network and the object.
- the pointer list shown in the lower part of FIG. 29 further associates a forward pointer and a backward pointer with a combination of an object identifier and a knowledge network identifier.
- the pointer is divided into two pointers, a forward pointer and a backward pointer.
- one pointer may be paired with information indicating the forward direction or the reverse method.
- These pointers represent the connection of things in the knowledge network.
- the pointer has an object identifier of another connected object.
- a pointer value that is not used may be a special value such as NULL.
- the pointer may also have a weight. This weight represents the degree of association with other connected objects. This weight may not be present.
- a node consists of a correspondence table between objects and knowledge networks and a pointer list.
- a pointer possessed by another object can be acquired.
- the node that is the object identifier pointed to by the pointer is acquired.
- acquiring the object identifier held by the pointer means that the object has been moved to the object specified by the pointer by tracing the pointer in the knowledge network. In other words, this means moving from one connected object to another in the knowledge network.
- Deleting the knowledge network identifier, the object identifier, and the pointer from the pointer list is deleting the connection between the objects.
- Adding knowledge network identifiers, object identifiers, and pointers to the pointer list is connecting objects.
- a set of attribute identifiers for identifying each of the plurality of attributes and the knowledge network identifier is created by the number of attributes, and these sets are added to the knowledge network / attribute correspondence table 101 shown in FIG. 28B.
- the attribute identifier of the attribute characterizing this knowledge network can be known.
- the knowledge network / attribute correspondence table 101 is searched using one attribute identifier as a key.
- the search result is searched using another attribute identifier as a key. By repeating this, the knowledge network identifier of the knowledge network characterized by all the attributes used for the search can be acquired.
- a knowledge network characterized by a plurality of attributes that is, a knowledge network in a specific field
- the attribute associated with the knowledge network may be selected from the attributes of all the objects in the knowledge network, or may be an attribute of some objects. Moreover, the attribute which the thing in the knowledge network does not have may be sufficient. Attributes specific to the knowledge network may be generated and associated with the knowledge network. If the correspondence table between the objects and the knowledge network in FIG. 29 is searched using the knowledge network identifier as a search key, the object identifiers of the objects belonging to the knowledge network can be acquired.
- an object identifier having the knowledge network identifier can be acquired by searching for the node using the knowledge network identifier as a key. In other words, you can see what is contained in the knowledge network.
- a pair of forward and backward pointers corresponds to an arrow between nodes.
- the forward pointer is used when moving to a node connected in the direction of the arrow.
- the backward pointer is used when moving to a node connected in the opposite direction to the arrow.
- a weight is held in the pointer, and the strength of the relation between the object directly associated with the pointer and the object having the pointer can be expressed.
- attribute hierarchy information It is one Embodiment of the hierarchy information of an attribute in the case of hierarchizing an attribute.
- This example is a data structure called a so-called list structure.
- An attribute identifier of one attribute holds a pointer to an upper layer and a pointer to a lower layer.
- the pointer to the attribute of the higher hierarchy holds the attribute identifier of the attribute of the higher hierarchy.
- the pointer to the lower hierarchy holds the attribute identifier of the attribute of the lower hierarchy. If the attribute does not have a pointer to the upper hierarchy or a pointer to the lower hierarchy, the pointer to the upper hierarchy or the lower hierarchy is a special value such as NULL indicating that there is no pointer.
- an attribute identifier is specified as a search key and a column of pointers to a higher hierarchy is searched, an attribute lower than the specified attribute is selected. If a column of pointers to lower layers is searched using the attribute identifier as a search key, an attribute higher than the specified attribute is selected. If the attribute identifier column is searched using the attribute identifier as a search key, a pointer to the upper hierarchy and a pointer to the lower hierarchy of the specified attribute are selected.
- a knowledge collection of graduates of a school is generated. Search the knowledge collection of graduates by Suzuki's identifier. If Mr. Suzuki's identifier can be selected, it can be determined that Mr. Suzuki is a graduate of this school. If a given thing is included in a knowledge collection generated by specifying one or more given attribute identifiers in advance, the given thing is a condition that is a given attribute. It is judged that
- Learning is to create or delete a knowledge network, add or delete an object to the knowledge network, add, delete, or change a connection between objects.
- the combination can be changed by combining addition and deletion. That is, when a connection that is no longer needed is deleted and a new connection is added, the connection is changed. It is also learning to add and delete attributes that are attached information of objects. It is also learning to add or delete feature data or identification data from the attribute.
- a knowledge network can be generated as follows.
- the knowledge network information 84 is added to the node 80 that is an object
- the object is added to the knowledge network.
- the knowledge network identifier and the object identifier are added to the correspondence table between the object and the knowledge network in FIG. 29 in association with each other. This adds knowledge network information to the node.
- the forward pointer 82 and the backward pointer 83 are added to the knowledge network information 84, an object can be connected. It is assumed that the two connected objects already have the knowledge network information 84. Specifically, if a forward direction pointer and a backward direction pointer are added to the pointer list of FIG. 29, the objects are connected. One object identifier and knowledge network identifier are associated with the forward pointer or backward pointer in which the other object identifier is set, and added to the pointer list. Similarly, the other object identifier and the knowledge network identifier are associated with the forward pointer or the backward pointer in which one object identifier is set, and added to the pointer list. This connects two things.
- the two objects are connected in both directions. If one of the former and the latter sets the other object identifier in the forward pointer, and the other sets one object identifier in the backward pointer, the two objects are connected in one direction.
- the two connected objects have pointers set with each other's object identifiers. If two pointers are deleted from the pointer list, the connection between the two objects can be deleted. If there is no pointer in the knowledge list for which an object identifier in the pointer list is set, the object may be deleted from the knowledge network. If the line that associates the knowledge network identifier with the object identifier is deleted from the correspondence table between the object and the knowledge network, the object is deleted from the knowledge network.
- the list of objects and attributes shown in FIG. 19 is searched using the object identifier and attribute identifier as keys. Deleting the selected line deletes the connection between the attribute and the object. It does not delete the attribute itself.
- An attribute can be added to an object by adding the correspondence between the object identifier and the attribute identifier to the list of objects and attributes.
- One attribute is extracted from the given attributes, and a list of objects and attributes is searched using the attribute identifier of the attribute as a key. Next, one attribute is extracted from the remaining given attributes, and the search result is further searched using the attribute identifier of the extracted attribute as a key. This process is repeated for all the given attributes, and when there is one kind of object identifier in the selected row, the object identifier represents an object having all the given attributes. The thing is identified by the given attribute.
- the search is repeated for all the given attributes, if there are a plurality of types of object identifiers, it can be seen that the plurality of object identifiers are candidates for the object to be identified. In this case, an attribute may be added and repeated until there is one kind of object identifier.
- Attribute is selected as follows. Feature data or identification data in the knowledge base determined to be the same as the feature data or identification data generated from the information input to the knowledge base system is selected and acquired. An attribute directly connected to and directly associated with the selected or acquired feature data or identification data is selected and acquired. Using this selected and acquired attribute, an object is recognized as described above.
- an object identifier of an object to be a starting point that is, a starting point node is designated (S141).
- the knowledge network identifier of the knowledge network to which the product identified by the product identifier belongs can be obtained.
- an object (node) associated with the object identified by the initially specified object identifier can be acquired (S143).
- the moved object identifiers are recorded in order, the recorded route becomes the travel path.
- ⁇ ⁇ ⁇ Move from one object to another connected object by following the pointer. There may be some relationship between the starting point and the destination. This process is also inference.
- the weights of the pointers are compared, and the pointer having the highest degree of association can be selected and moved.
- the weight of each knowledge network information can be compared to move to a knowledge network with the least degree of association.
- the word “Dog” in the language is input to the knowledge base system by voice, and identification data is generated and stored in the knowledge base of one embodiment.
- the feature extraction unit 1020 When the word “Dog” is input by voice, the feature extraction unit 1020 generates identification data of the voice data.
- the feature extraction unit 1020 when a character image “Dog” as a word is input, the feature extraction unit 1020 generates image identification data. Further, when “Dog” is input as the character code of the word, the feature extraction unit 1020 generates character code identification data.
- “Dog” is represented by a phonetic symbol, but the expression format of the identification data is not limited to this, and may be the above-described audio data, video data, or character code.
- the expression format of the identification data is not limited to this, and may be the above-described audio data, video data, or character code.
- binary data representing the signal intensity for each frequency component, such as a sound intensity of 100 Hertz being 55, a sound intensity of 200 Hertz being 156.
- Pattern data in which numerical values are arranged becomes identification data.
- the feature data and the identification data represented by the note marks in FIG. 33 and the like are the above-described sound data or sound pattern data.
- identification data is not used as a direct representation of meaning or concept.
- the feature data directly holds and represents the meaning and concept. Therefore, the meaning and concept cannot be understood unless the feature data is acquired.
- the knowledge base of one embodiment includes objects and attributes.
- an object (3003) shown in FIG. 33 represents a dog.
- the attribute (1003) represents the feature of the appearance of the dog.
- “thing (xxxx)” refers to an entity identifier “xxxx”
- “attribute (yyyy)” refers to an attribute identifier “yyyy”.
- a video of a dog is taken by a camera, and the video data is input to a knowledge base system 1000 which is an embodiment of the present technology.
- a camera as an input device captures a video of a dog, and the captured video data is input to the knowledge base system 1000 by the data input unit 1010.
- Video data that is input data received by the data input unit 1010 is passed to the feature extraction unit 1020.
- the feature extraction unit 1020 extracts features from input data and generates feature data.
- the generated feature data is passed to the data comparison unit 1030.
- the data comparison unit 1030 compares the already stored feature data with the passed feature data to check whether there is the same feature data.
- the data storage unit 1040 stores the passed feature data in the knowledge base, generates a new attribute in the knowledge base, and stores the stored feature data and the newly generated attribute. connect.
- the data storage unit 1040 generates a new object in the knowledge base, and connects the newly generated object and the generated attribute.
- an object that is indirectly associated with the dog that is the object that directly emitted the video is generated in the knowledge base.
- Objects in the knowledge base are not directly associated with objects.
- a dog's call “bow” is input to the knowledge base system 1000 from a microphone while a video of a dog, which is an object, is being shot by a camera.
- the video of the puppy that is the object is taken by the camera and input to the data input unit 1010 of the knowledge base system 1000.
- the video data input to the data input unit 1010 is passed to the feature extraction unit 1020.
- the feature extraction unit 1020 generates feature data from the input video data.
- the data comparison unit 1030 checks whether the same feature data is already in the knowledge base. Since there is no same feature data, the data storage unit 1040 stores the feature data in the knowledge base, generates the attribute (1021) of FIG. 33, and connects the stored feature data to the attribute (1021).
- the data storage unit 1040 generates the object (3003) of FIG. 33 and connects the attribute (1021) and the object (3003).
- the knowledge base system 1000 receives information from outside the knowledge base system 1000, generates feature data from the received information, and stores the generated feature data in the knowledge base.
- An attribute is generated and the stored feature data is linked to that attribute.
- an object is generated, and the generated attribute is connected to the generated object. It is learning by addition of an object that feature data is connected to a newly generated object via an attribute.
- ⁇ ⁇ Input processing is performed in the same way for dog calls input from microphones.
- the voice data “dogwow” of the dog input to the data input unit 1010 through the microphone is passed to the feature extraction unit 1020.
- the feature extraction unit 1020 generates feature data from the acquired voice data.
- the data comparison unit 1030 determines that the same feature data is not in the knowledge base.
- the generated feature data is stored in the knowledge base by the data storage unit 1040.
- the data storage unit 1040 generates the attribute (1011) in FIG. 33 and connects the stored feature data to the attribute (1011).
- an object may be generated and an attribute (1011) may be connected to the generated object. Of course, this product need not be generated.
- the learning unit 1050 performs the following learning process.
- the following learning process may be automatically started.
- the learning process may be instructed in advance by performing keyboard input or mouse operation at the start of input in step 1.
- the learning unit 1050 acquires feature data in the knowledge base that is first generated and stored, or feature data in the knowledge base that is determined to be the same as the feature data that is generated first.
- the search unit 1060 acquires an attribute (1021) connected to the acquired feature data. Furthermore, the search unit 1060 acquires an object (3003) connected to the acquired attribute (1021). The learning unit 1050 acquires the acquired object (3003) and recognizes the object (3003).
- the learning unit 1050 acquires feature data generated from the dog's bark “bow” and stored later, or feature data in the knowledge base determined to be the same as the feature data generated later.
- the search unit 1060 acquires an attribute (1011) connected to the feature data.
- the two attributes (1021, 1011) are not directly connected to the same object, so the learning unit 1050 performs the following process.
- the learning unit 1050 connects the recognized object (3003) and the attribute (1011) that is generated later and is not directly connected to the recognized object (3003).
- the thing (3003) has the characteristic data of the dog's cry through the attribute (1011). Step 1 in FIG. 33 is in this state.
- the standard for determining whether or not data is input continuously is held in the knowledge base system 1000 in advance.
- a new standard can be received from outside the knowledge base system 1000, the standard can be maintained, and the standard can be additionally changed.
- a criterion for determining whether or not they are the same is also stored in the knowledge base system 1000 in advance.
- a new standard can be received from outside the knowledge base system 1000, the standard can be maintained, and the standard can be additionally changed.
- the feature data is stored in the knowledge base and the knowledge base is constructed as described above, the following recognition process can be performed.
- a dog cry is input to the data input unit 1010 of the knowledge base system 1000 through a microphone.
- the voice data of the dog cry is passed from the data input unit 1010 to the feature extraction unit 1020, and the feature extraction unit 1020 generates feature data from the delivered voice data.
- the data comparison unit 1030 receives the generated feature data and checks whether the same feature data exists in the knowledge base.
- the search unit 1060 acquires the attribute (1011) connected to the found feature data. Further, the search unit 1060 acquires and recognizes an object (3003) connected to the attribute (1011).
- the search unit 1060 can acquire another attribute (1021) connected to the object (3003), and can acquire other feature data generated from the video of the dog connected to the attribute (1021).
- the acquired feature data is feature data generated from information such as an image directly emitted from a recognized object. In this way, other feature data other than the call used to recognize the object (3003) can be acquired.
- the output unit 1080 displays the video of the dog on the display device that is the output device based on the feature data.
- the object can be recognized by the cry and the image of the dog that is the recognized object can be output.
- the dog's call “bow” is input to the knowledge base system 1000 from the microphone.
- the voice data input to the data input unit 1010 through the microphone is passed to the feature extraction unit 1020 to generate feature data.
- the data comparison unit 1030 receives the generated feature data and checks whether the same feature data exists in the knowledge base.
- the search unit 1060 acquires an attribute (1011) connected to the found feature data. Further, the search unit 1060 acquires an object (3003) connected to the attribute (1011). Through the above processing, the object (3003) is recognized by the dog's cry “bowwow” input with the microphone.
- the judgment criteria for judging whether or not they are the same are provided in the knowledge base system 1000 in advance.
- a new criterion can be acquired from outside the knowledge base system 1000 and held.
- Judgment criteria for judging whether or not they are the same are called the same criteria.
- the video data input to the data input unit 1010 is passed to the feature extraction unit 1020.
- the feature extraction unit 1020 generates feature data from the input video data.
- the data comparison unit 1030 that has received the generated feature data checks whether the same feature data exists in the knowledge base. In this case, it is determined that there is no same feature data.
- the data storage unit 1040 stores the generated feature data in the knowledge base, generates the attribute (1022) in FIG. 33, and connects the stored feature data to the generated attribute (1022).
- an object may be generated by the data storage unit 1040, and the generated object may be connected to the attribute (1022). This product need not be generated.
- the learning unit 1050 performs the following learning process.
- the following learning process may be automatically started. Learning processing may be instructed in advance by performing keyboard input or mouse operation before input.
- the learning unit 1050 obtains feature data in the knowledge base that is generated from the dog's call “bowwow” and is determined to be the same as the feature data that was first passed.
- the search unit 1060 acquires an attribute (1011) connected to the acquired feature data. Further, the search unit 1060 acquires an object (3003) connected to the acquired attribute (1011). The learning unit 1050 acquires the acquired object (3003) and recognizes the object (3003).
- the learning unit 1050 obtains feature data that is generated from a large dog video, passed later, and stored.
- the search unit 1060 acquires an attribute (1022) connected to the feature data. Since the two acquired attributes (1011 and 1022) are not connected to the same object, the learning unit 1050 performs the following process.
- the learning unit 1050 connects the recognized object (3003) and the attribute (1022) acquired later and not directly connected to the recognized object (3003).
- An object (3003) has characteristic data generated from a video of a large dog via only the attribute (1022).
- This learning process is an integration process of feature data stored in the knowledge base, which is performed after the learning process by adding the feature data in Step 2. This is a process for connecting feature data connected to two different attributes to one common attribute.
- the learning unit 1050 performs the following integration processing. It may be instructed by mouse operation or keyboard input to perform integration processing at the start of step 2.
- the video of a small dog is input from the camera, and the feature data is generated by the feature extraction unit 1020.
- the data comparison unit 1030 checks whether the same feature data as the generated feature data exists in the knowledge base. Since the same feature data exists in the knowledge base, the attribute (1021) connected to the feature data is acquired by the search unit 1060. Further, an object (3003) connected to the acquired attribute (1021) is acquired by the search unit 1060.
- the data comparison unit 1030 checks whether the same feature data as the generated feature data exists in the knowledge base. Since the same feature data exists in the knowledge base, the attribute (1022) connected to the feature data is acquired by the search unit 1060. Further, an object (3003) connected to the acquired attribute (1022) is acquired by the search unit 1060.
- the learning unit 1050 performs the following process.
- the learning unit 1050 acquires two attributes (1021, 1022). When the number of feature data connected to two different attributes is one, the learning unit 1050 performs the following process. A new attribute (1003) is generated, the two feature data are connected to the generated attribute (1003), and the connection of the object (3003) with the two attributes (1021, 1022) is deleted. Without generating a new attribute (1003), one of the two attributes (1021, 1022) is directly connected to the feature data connected to the other attribute, and the other attribute is connected to the object (3003). May be deleted.
- the learning unit 1050 When only the number of feature data connected to one attribute is 1 and only the number of feature data connected to the other attribute is greater than 1, the learning unit 1050 performs the following process.
- the feature data of the attribute to which only one feature data is connected is connected to the other attribute, and the connection of the object with the attribute to which only one feature data is connected is deleted.
- feature data When the feature data is newly stored in the knowledge base, an attribute is generated, and the generated attribute is connected to the stored feature data.
- feature data When feature data is stored in the knowledge base, only one feature data is connected to an attribute connected to the feature data. Therefore, feature data can be aggregated as described above.
- Attribute that multiple feature data is connected is a bundle of several feature data connected to the attribute, and represents the nature, feature, meaning or concept of the aggregated feature data.
- the recognition process can be performed even in the state of step 3.
- feature data is generated from the input dog bark.
- the generated feature data is compared with the feature data in the knowledge base.
- the attribute (1011) connected to the found feature data is acquired.
- the object (3003) connected to the acquired attribute (1011) is acquired, if only one other attribute (1003) connected to the object (3003) is acquired, a plurality (two in this example) of feature data Can be obtained.
- the two feature data cannot be acquired unless the two attributes (1021) and (1022) are acquired.
- Feature data is generated from the dog cry input by the feature extraction unit 1020, and the generated feature data is passed to the data comparison unit 1030. Since the feature data generated from the dog's bark is already stored in the knowledge base, the data comparison unit 1030 determines that there is the same feature data. Then, the search unit 1060 acquires the attribute (1011) connected to the feature data determined to be the same. Further, the search unit 1060 acquires an object (3003) connected to the attribute (1011). And if the other attribute (1003) connected with the thing (3003) is acquired, the feature data connected with the attribute (1003) can be acquired. Based on the acquired feature data, the output unit 1080 displays the image of the dog on the display device. Through this series of processing, it is recognized that the dog is a dog from the cry of the dog, and the image of the recognized dog is displayed.
- step 2 In the state of step 2, two attributes of the attribute (1021) and the attribute (1022) are acquired, but in the state of step 3, if one attribute (1003) is acquired, two feature data are acquired. Can be obtained. Since the number of times of acquiring the attribute connected to the object is reduced, the feature data can be acquired faster than Step 2.
- the video of the puppy is input from the camera to the data input unit 1010 of the knowledge base system 1000.
- Feature data is generated from the video input by the feature extraction unit 1020, and it is checked whether the same feature data as the feature data generated by the data comparison unit 1030 exists in the knowledge base. The same feature data is found, and the attribute (1003) to which the same feature data is connected is acquired by the search unit 1060. Further, the search unit 1060 acquires an object (3003) connected to the acquired attribute (1003). As described above, it is recognized from the video of the puppy input from the camera that the object directly emitting the video is the object (3003).
- the word “Dog” is input to the data input unit 1010 of the knowledge base system 1000 from the microphone.
- Identification data is generated from the voice input by the feature extraction unit 1020, and it is checked whether the same identification data as the identification data generated by the data comparison unit 1030 exists in the knowledge base. Since the same identification data cannot be found, the generated identification data is stored in the knowledge base by the data storage unit 1040.
- the data storage unit 1040 further generates an attribute (1013), and connects the generated attribute (1013) and the stored identification data.
- an object may be generated by the data storage unit 1040, and the attribute (1013) stored in the generated object may be connected. This product need not be generated.
- the learning unit 1050 performs the following learning process.
- the following learning process may be automatically started. It may be instructed by mouse operation or keyboard input to perform integration processing at the start of step 4.
- the learning unit 1050 acquires feature data in the knowledge base that is generated from the video of the puppy and is determined to be the same as the feature data that was first passed.
- the search unit 1060 acquires an attribute (1003) connected to the acquired feature data. Further, the search unit 1060 acquires an object (3003) connected to the acquired attribute (1003).
- the learning unit 1050 acquires the acquired object (3003) and recognizes the object (3003).
- the learning unit 1050 acquires identification data that is generated from the voice of the word “Dog”, passed later, and stored.
- the search unit 1060 acquires an attribute (1013) connected to the identification data. Since the two acquired attributes (1003, 1013) are not connected to the same object, the learning unit 1050 performs the following process.
- the learning unit 1050 connects the recognized object (3003) and the attribute (1013) acquired later and not directly connected to the recognized object (3003).
- the object (3003) has the identification data generated from the word “Dog” via the attribute (1013).
- identification data generated from the speech of the word “Dogy” is stored in the knowledge base.
- An attribute (1014) is generated and linked to the stored identification data.
- the generated attribute (1014) is connected to the object (3003).
- the video of the puppy and the word “Doggy” may be associated with each other, or the word “Dog” and the word “Doggy” may be associated with each other.
- This learning process is an integration process of identification data stored in the knowledge base, which is performed after the learning process by adding the identification data in Step 4. This is a process of connecting identification data connected to two different attributes to one common attribute.
- the learning unit 1050 performs the following integration process. It may be instructed by mouse operation or keyboard input to perform integration processing at the start of step 4.
- the voice of the word “Dog” is input from the microphone, and the identification data is generated by the feature extraction unit 1020.
- the data comparison unit 1030 checks whether the same identification data as the generated identification data exists in the knowledge base. Since the same identification data exists in the knowledge base, an attribute (1013) connected to the identification data is acquired by the search unit 1060. Further, an object (3003) connected to the attribute (1013) is acquired by the search unit 1060.
- the voice of the word “Dogy” is input from the microphone, and the feature extraction unit 1020 generates the identification data.
- the data comparison unit 1030 checks whether the same identification data as the generated identification data exists in the knowledge base. Since the same identification data exists in the knowledge base, the attribute (1014) connected to the identification data is acquired by the search unit 1060. Further, the search unit 1060 acquires an object (3003) connected to the attribute (1014).
- the learning unit 1050 performs the following process.
- Learning unit 1050 acquires two attributes (1013, 1014). When the number of identification data connected to two different attributes is one, the learning unit 1050 performs the following process. Create a new attribute, connect the two identification data to the generated attribute, connect the generated attribute to the object (3003), and delete the connection between the two attributes (1013, 1014) and the object (3003) To do. Or, without generating a new attribute, one of the two attributes (1013, 1014), for example, the attribute (1013) is connected to the identification data connected to the other attribute (1014), and the other attribute The connection between (1014) and the object (3003) may be deleted. In order to make the figure easier to see, the attribute (1014), the attribute (1021), and the attribute (1022) are omitted in the diagram of step 5.
- the learning unit 1050 When only the number of identification data connected to one attribute is 1 and only the number of identification data connected to the other attribute is larger than 1, the learning unit 1050 performs the following process.
- the identification data of the attribute connected to only one identification data is connected to the other attribute, and the connection between the attribute connected to only one identification data and the object is deleted.
- An attribute to which a plurality of identification data is connected represents a collection of identification data that is aggregated by bundling together several identification data connected to the attribute.
- the feature extraction unit 1020 extracts the feature and generates identification data.
- the data comparison unit 1030 checks whether the same identification data as the generated identification data exists in the knowledge base. Since the same identification data exists in the knowledge base, the search unit 1060 acquires an attribute (1013) connected to the identification data. Furthermore, the search unit 1060 acquires an object (3003) connected to the acquired attribute (1013).
- the search unit 1060 acquires another attribute (1003) connected to the object (3003), and acquires feature data connected to the attribute (1003).
- the output unit 1080 displays a dog image on the display device based on the acquired feature data. Further, the search unit 1060 acquires another attribute (1011) connected to the object (3003), and acquires feature data connected to the attribute (1011). Based on the acquired feature data, the output unit 1080 outputs the dog's call “bow” from the voice reproduction device which is an output device. The dog is recognized from the word “Dog”, and the video of the dog and the cry of the dog are output.
- identification data is not necessary.
- An animal that does not have a language is considered to construct a knowledge base that does not include identification data and perform knowledge processing such as recognition and inference using the knowledge base that does not include identification data.
- a person who has learned a language is thought to have built a knowledge base that includes both identification data and feature data.
- (1) Knowledge base made up of feature data, attributes and objects without identification data (2) Knowledge base made up of identification data, attributes and things without feature data, (3) Identification data, feature data
- a knowledge base made up of attributes and objects can be constructed. When performing only symbolic logic, the knowledge base of (2) may be used.
- feature data, identification data, attributes, and objects in the knowledge base may be added or deleted.
- feature data, identification data, attributes, and connection of objects are added or deleted and changed.
- identification data in the knowledge base of one embodiment is generated from only words in a language, and only from data such as video and audio directly emitted from objects in the real world, Feature data in the knowledge base of one embodiment is generated.
- the object is not a word.
- identification data is not used as a direct representation of meaning or concept.
- the attribute connected to the identification data is acquired, the object connected to the attribute is acquired, the other attribute connected to the acquired object is acquired, and the attribute is acquired.
- the object (3003) is indirectly associated with the object called a dog.
- the object (3003) may be indirectly associated with an object called a cat instead of an object called a dog. Therefore, by generating identification data from words in the language, connecting the identification data with objects and attributes in the knowledge base, and associating them, information can be exchanged between different knowledge bases via words. Yes.
- the identification data is used to associate a language with a knowledge base outside the language.
- Different knowledge bases have different identifier values assigned to the same objects and attributes, so information cannot be exchanged without using identifiers generated from words in the language.
- the identification data can also be used to extract knowledge in the knowledge base into the language.
- Knowledge in the knowledge base can be extracted from the knowledge base into the language by converting the identification data into words using the intervening data. For example, if an object represented by an identifier in the knowledge base is converted into a word with identification data as an intermediary, it becomes easier to distinguish the object than distinguishing an object only by the identifier.
- an object identifier or attribute identifier in the knowledge base is converted into a word, it can be written in characters. And it becomes easy to understand by recognizing the written thing and attribute as a word using vision.
- New feature data may be generated from one stored feature data.
- feature data of four legs is generated and stored in the knowledge base. This is further generated by the feature extraction unit 1020 from feature data generated from already stored videos of dogs, cats, lizards, and the like.
- the knowledge base of FIG. 34 includes an object representing a human (3001), an object representing a dolphin (3002), an object representing a dog (3003), an object representing a cat (3004), an object representing a lizard (3005), and a snake.
- An object to be represented (3006) is stored.
- FIG. 35 shows objects and attributes newly generated by the learning process performed by the learning unit 1050 based on the knowledge base of FIG.
- the voice of the word “mammal” is input to the data input unit 1010 of the knowledge base system 1000 using a microphone.
- Features are extracted by the feature extraction unit 1020, and identification data is generated.
- the data comparison unit 1030 checks that the identification data is not in the knowledge base, and the data storage unit 1040 stores the identification data in the knowledge base.
- the attribute (1015) is generated by the data storage unit 1040, and the generated identification data is connected to the generated attribute (1015).
- an object (3007) is generated by the data storage unit 1040, and the generated attribute (1015) is connected.
- Attribute integration processing is instructed by mouse operation or keyboard input.
- the voice of the word “mammal” is input to the data input unit 1010 of the knowledge base system 1000 with the microphone.
- Features are extracted by the feature extraction unit 1020, and identification data is generated.
- the data comparison unit 1030 checks whether the identification data is in the knowledge base.
- the search unit 1060 acquires an attribute (1015) connected to the same identification data, and acquires an object (3007) connected to the acquired attribute (1015).
- the learning unit 1050 acquires and recognizes the acquired object (3007).
- the learning unit 1050 When an instruction to generate a new attribute is input to the knowledge base system 1000 by mouse operation or keyboard input, the learning unit 1050 generates the attribute (1008), and the generated attribute (1008) is recognized (3007). ) And the generated attribute (1008) is acquired. A list of attributes already connected to the recognized object (3007) is displayed on the display device, and one attribute can be specified from the list to acquire the specified attribute.
- a video of “big dog” is input from the camera to the data input unit 1010 of the knowledge base system 1000, and features are extracted by the feature extraction unit 1020 to generate feature data.
- the data comparison unit 1030 checks whether the same feature data as the generated feature data exists in the knowledge base. Since there is the same feature data, the search unit 1060 acquires the same feature data in the knowledge base.
- the acquired feature data is connected to the acquired attribute (1008) by the learning unit 1050.
- the attribute (1008) is connected to a plurality of feature data generated from the image of the mammal, and the attribute (1008) represents the meaning or concept of a mammal.
- This attribute (1008) leads to an object (3007) representing a mammal.
- An object representing an animal (3008) and an object representing a reptile (3009) are produced in the same manner.
- FIG. 36 shows a knowledge base in which the three attributes (1008, 1009, 1010) of FIG. 35 are added to FIG. 34 and the attribute (1008) is connected to the thing (3001, 3002, 3003, 3004).
- an attribute (1008) representing a mammal is added to the object (3001, 3002, 3003, 3004), and the attribute (1008) is obtained. Similar processing is performed for the attributes (1009, 1010).
- Attribute addition processing by the learning unit 1050 is performed as follows.
- the voice of the word “mammal” is input from the microphone to the data input unit 1010 of the knowledge base system 1000.
- Features are extracted by the feature extraction unit 1020, and identification data is generated.
- the data comparison unit 1030 confirms that the same identification data as this identification data exists in the knowledge base.
- the attribute (1015) connected to the same identification data is acquired by the search unit 1060.
- an object (3007) connected to the attribute (1015) is acquired and recognized by the search unit (1060).
- “human” video is input from the camera, feature data is generated, and an attribute (1001) to which the same feature data as the generated feature data is connected is acquired. Further, an object (3001) connected to the attribute (1001) is acquired and recognized by the search unit 1060. If the first recognized object (3007) is different from the next recognized object (3001), the learning unit 1050 determines that the attribute (1008) directly connected to the first recognized object (3007) and the later Connected to the recognized object (3001).
- an object representing a human (3001), an object representing a dolphin (3002), an object representing a dog (3003), and an object representing a cat (3004) are connected to an attribute representing a mammal (1008).
- objects connected to the attribute (1008) can be selected, and objects having the attribute (1008) can be collected.
- Objects (3001, 3002, 3003, 3004) are selected and collected. This collected object is called a knowledge collection.
- Objects having the attribute of mammal (1008) are collected, and a collection of knowledge of mammals is generated.
- FIG. 37 shows four knowledge collections: animals, mammals, reptiles, and quadruped animals.
- a knowledge collection 166 For example, inferring "whether snakes are mammals”. Collecting things connected to the attribute (1008) representing mammals, a knowledge collection 166 is created.
- This knowledge collection 166 includes an object representing a person (3001), an object representing a dolphin (3002), an object representing a dog (3003), and an object representing a cat (3004). It is checked whether or not there is an object (3006) representing a snake in this knowledge collection. In this example, since the object representing the snake (3006) is not included in the knowledge collection representing the mammal, it can be inferred that "the snake is not a mammal".
- a collection of knowledge is an abstract object. Therefore, as with a specific object, a knowledge collection having an abstract knowledge collection as an element can also be generated. Add the same attribute to several abstract objects. Collecting objects with added attributes creates a new collection of knowledge.
- an abstract object (3007, 3008, 3009) representing a collection of knowledge newly generated in FIG. 35 is added and directly connected to the corresponding attribute.
- a collection of knowledge is stored in the knowledge base as an abstract object.
- an abstract object (3009) representing a reptile is connected to an attribute (1010).
- This attribute (1010) represents the reptile by connecting the feature data generated from the snake image and the feature data generated from the lizard image.
- an abstract object (3009) called a reptile and a specific object such as a snake or a lizard (3005, 3006) constituting the reptile are represented by the same structure in the knowledge base.
- feature data and attributes are connected, and the attributes are connected to objects. Therefore, it is possible to generate a knowledge collection whose elements are abstract objects called knowledge collections.
- objects are connected by arrows.
- an object representing an mammal (3007) An arrow extending to an object (3001), an arrow extending from an object representing a mammal (3007) to an object representing a dolphin (3002), an arrow extending from an object representing a mammal (3007) to an object representing a dog (3003), an arrow representing a mammal
- Objects are connected to each other by an arrow extending to.
- FIG. 40 is a diagram in which only the connection between objects is extracted.
- Several such connected objects that is, an entire connected object is a knowledge network.
- this knowledge network represents the classification of animal species.
- the start point of the arrow represents the upper classification, and the end point of the arrow represents the lower classification.
- This knowledge network connection represents an inclusive relationship, and an object located at the upper level includes an object located at the lower level.
- One knowledge network in FIG. 40 can be created by combining a plurality of knowledge networks.
- the object 41 is connected to an attribute (1030) representing a knowledge network and an attribute (1009) representing an animal, and is an abstract object that is an animal knowledge network.
- the object (3012) is connected to the attribute (1030) representing the knowledge network and the attribute (1008) representing the mammal, and is an abstract object that is a knowledge network about the mammal.
- the object (3013) is connected to the attribute (1030) representing the knowledge network and the attribute (1010) representing the reptile, and is an abstract object that is a knowledge network regarding the reptile.
- FIG. 42 is a diagram in which connections between the objects in FIG. 41 are extracted.
- a connection between an object that is a knowledge network and a specific object that is omitted in FIG. 41 for example, a connection between an object (3011) that is an animal knowledge network and an object (3008) that represents an animal.
- an object (3011) that is an animal knowledge network for example, a connection between an object (3011) that is an animal knowledge network and an object (3008) that represents an animal.
- FIG. 41 since the figure becomes complicated and difficult to see, it is omitted.
- This knowledge network is an abstract object (3011) which is a knowledge network related to animals, and is a portion surrounded by an ellipse 169 in FIG.
- the thing (3007) which is a mammal and the thing (3001) which is a person are connected
- the thing (3007) which is a mammal and the thing (3002) which is a dolphin are connected
- the thing (3007) which is a mammal and a dog are connected
- a thing (3003) is connected
- a thing (3007) that is a mammal and a thing (3004) that is a cat are connected
- a knowledge network is generated.
- This knowledge network is an abstract object (3012) which is a knowledge network related to mammals, and is a portion surrounded by an ellipse 170 in FIG.
- This knowledge network is an abstract object (3013) which is a knowledge network related to reptiles, and is a portion surrounded by an ellipse 171 in FIG.
- an abstract entity (3011) that is a knowledge network about animals is connected to an abstract entity (3012) that is a knowledge network about mammals, and an abstract that is a knowledge network about animals.
- a general object (3011) and an abstract object (3013) which is a knowledge network about reptiles are connected to generate a knowledge network. Note that this knowledge network is not drawn in FIG. 42 because it is difficult to see. In this way, a new knowledge network can be generated by connecting abstract objects that are knowledge networks.
- An abstract object that is a knowledge network is directly connected to each object included in the knowledge network.
- an abstract object (3011) which is a knowledge network is directly connected to an object (3007, 3008, 3009) included in the knowledge network.
- an abstract object that is a knowledge network is designated, it is possible to know what is included in the knowledge network.
- a collection of knowledge is a thing, and a knowledge network is also a thing.
- the data structure is the same for specific objects such as dogs and cats and abstract objects such as knowledge collections and knowledge networks. Therefore, it is possible to generate a knowledge network including a collection of knowledge that is an abstract object or a knowledge network that is an abstract object.
- a collection of knowledge can be generated that includes a collection of knowledge that is abstract or a knowledge network that is abstract.
- Abstract knowledge is also accumulated by this mechanism, and the accumulated abstract knowledge is systematized. In addition, inferences are made using accumulated and systematic abstract knowledge.
- a knowledge network (3012) about mammals it is possible to move from a human thing (3001) to a mammal thing (3007).
- the thing (3007) which is a mammal belongs to not only the knowledge network (3012) regarding a mammal but the knowledge network (3011) related to an animal.
- the animal moves to the animal knowledge network (3011) which is a different knowledge network from the mammal knowledge network (3012) via the mammal object (3007).
- the object moves from the mammal (3007) to the animal (3008).
- connection changes, make that part another knowledge network. Within one knowledge network, the meaning of the connection is the same. Then, the meaning of the knowledge network becomes clear. Further, if the attributes characterizing the knowledge network represent the meaning of the connection, knowledge networks having different connection meanings are always characterized by combinations of different attributes.
- the knowledge network can also be used as a search system for searching for related objects.
- tacit knowledge There is a conventional knowledge base that expresses knowledge by connecting words in a language. It connects words to create a single connection and expresses knowledge. However, it was not possible to express all knowledge with a lump of words. Therefore, knowledge that could not fit into a single block of words was called tacit knowledge.
- tacit knowledge For example, if a person belongs to a company organization, there is a connection in the company organization. There is a family relationship at home. Everything, not just people, has various connections depending on the cut. It is impossible to express this with a lump of connections. Forcibly expressed as a group of connections, knowledge or information that cannot fit into a group of connections must be called tacit knowledge.
- a lump of connections is made for each cut.
- This group of connections is a knowledge network.
- knowledge can be organized by connecting knowledge networks with common objects, and a wide range of knowledge can be organized without using tacit knowledge. You can move from one object belonging to a certain knowledge network to another knowledge network connected to that knowledge network, move to another object in the other knowledge network, and check the connection with the object in the other knowledge network. .
- the brain is made up of many nerve cells. Furthermore, nerve cells are connected to each other by tissues called dendrites and transmit signals. In the actual brain, one nerve cell corresponds to an object.
- information directly generated from an external object is received by the sensory organ 110.
- Features are extracted from the information received by the sensory organ 110 by the feature extraction unit 114, and feature data is generated.
- the generated feature data is held in the storage area 112.
- the feature data is connected to one nerve cell 71 in the attribute region 115 by a dendrite or the like, and the feature data is associated with the nerve cell 71 in the attribute region 115.
- a plurality of feature data may be connected to the same nerve cell 71 in the attribute region 115 and associated therewith.
- the nerve cell 71 in the attribute area 115 represents an attribute by being associated with feature data.
- the nerve cell 71 representing the attribute and the nerve cell 70 in the object region 111 are connected by a dendrite or the like, and the attribute is associated with the object. In this way, things have attributes.
- the feature extraction unit 114 extracts features from the sweetness information, and the extracted features are held in the storage area 112 and become feature data.
- the feature data is connected to the nerve cell 71 in the attribute region 115, the attribute that the nerve cell is sweet is represented.
- the nerve cell 71 representing the attribute of sweetness is directly associated with the object and directly connected to the nerve cell 70 representing the object. Thereby, it is learned that “the object is sweet” or “the object has a sweet attribute”.
- one feature data can be connected to a plurality of different attributes.
- a plurality of feature data can be connected to one attribute to generate one meaning or concept.
- one meaning and concept can be generated from a plurality of feature data, or a plurality of meanings and concepts can be generated from one feature data.
- the feature data may be associated with the nerve cell 70 in the object area 111 without passing through the nerve cell 71 in the attribute area 115. In this case, the signal reaches the nerve cell 70 in the object area 111 faster than the nerve cell 71 in the attribute area 115.
- information is received from the sensory organ 110 and it is determined that the feature data extracted from the information received by the feature extraction unit 114 is identical or similar to the feature data already stored in the storage area 112. If the nerve cell 70 in the object region 111 and its feature data are directly connected, a signal reaches the nerve cell 70 in the object region 111 connected to the feature data in advance. At this time, the object is recognized.
- a mechanism that immediately recognizes a fire when viewing the image of a flame burning in red is associated with the aforementioned mechanism.
- a red round object is eaten and sweet.
- one neuron 70 in the object region 111 becomes a red round object, and it is connected to the feature data of the sweetness of the nerve cell and the red round object, which is the attribute connected to the feature data of the image of the red round object.
- the nerve cells 71 in the attribute region 115 are connected to the red round nerve cells.
- the feature data that can be eaten is connected to the nerve cell 71 in the attribute region 115, and the nerve cell 71 in the attribute region 115 is connected to the nerve cell 70 corresponding to the red round object in the object region 111.
- the nerve cell 71 in the attribute area 115 connected to the feature data that can be eaten sends a signal to the nerve cell 70 in the object area 111. That is, when eating, a signal is sent from the nerve cell 71 in the attribute area 115 to the nerve cell 70 in the object area 111. Then, only the nerve cell 70 in the region 111 of the object connected to the nerve cell 71 in the attribute region 115 that can be eaten receives the signal and is activated. The entire nerve cell 70 in the activated object region 111 becomes a collection of objects having an attribute of being eaten. This gathering becomes a knowledge gathering. It is also possible to send a signal from the neuron 71 of one or more attribute regions 115 to generate a knowledge collection.
- the nerve cell 70 in the object region 111 represents an object.
- the nerve cells in the knowledge collection represent things with common attributes. By connecting nerve cells in one knowledge collection, the relationship between objects can be maintained.
- a collection of connected objects is a knowledge network. Of course, you can also connect things that are not in the knowledge collection. Furthermore, a knowledge network may be created by connecting objects regardless of the knowledge collection.
- Adding or deleting object, attribute or feature data is learning.
- Learning to add or delete connections is also learning.
- adding connections between objects deleting connections between objects, adding new neurons to the knowledge network in association with objects, objects and attributes, attributes and feature data, feature data and objects Learning to add and delete connections with.
- the feature extraction unit 114 extracts the above two types of information as features.
- the two extracted features are held in the storage area 112 as a set of feature data.
- the same feature data already exists in the storage area 112 it is possible not to newly store the feature data.
- one of the unused nerve cells in the attribute area 115 is connected to the feature data and becomes a nerve cell corresponding to the feature data.
- one of the nerve cells in the attribute region 115 is connected to the feature data and becomes a nerve cell corresponding to the feature data.
- the feature data of sweetness and the nerve cells in the attribute area 115 are connected, and the feature data of the degree of sweetness and another nerve cell in the attribute area 115 are connected.
- An attribute representing the degree of sweetness can be linked to another attribute as an attribute representing the degree.
- Auxiliary attributes connect to other attributes and modify the meaning and concept of the attribute. Auxiliary attributes are easy to understand if they are considered to correspond to adjectives and adverbs.
- a nerve cell associated with an object that is a source of information also receives a signal.
- the two nerve cells in the attribute region 115 that have received the signal are each connected to a nerve cell corresponding to an object that is a source of information.
- Objects, attributes, and auxiliary attributes are all connected to form a triangle. If two of the three elements are specified, the remaining one element is always determined. For example, if an object and an attribute are specified, only one auxiliary attribute connected to both is determined. In this way, auxiliary attributes related to the attribute can be acquired when necessary. Of course, if the feature data connected to the attribute is acquired, the information that is the source of the auxiliary attribute can be acquired, so that more detailed information can be acquired.
- the auxiliary attribute is also an attribute, and is merely connected to the feature data and the nerve cell in the attribute region 115. One auxiliary attribute is connected to various attributes, and the meaning and concept of the connected attribute can be modified.
- Feature data corresponding to sweetness is a main feature, and feature data corresponding to the degree of sweetness is a subfeature.
- the attribute corresponding to the main feature is the main attribute
- the attribute corresponding to the sub feature is the sub attribute.
- the attribute called the auxiliary attribute is the sub attribute.
- the feature extraction unit 114 When the feature extraction unit 114 receives information from the sensory organ 110, a main feature and sub features are extracted as necessary. The extracted main feature and sub feature are stored in the storage area 112 as feature data.
- the feature data that is the main feature of sweetness is connected to the nerve cell that represents the main attribute of sweetness
- the feature data that is the subfeature of the degree of sweetness is connected to the nerve cell that represents the subattribute corresponding to the subfeature of the degree of sweetness.
- a nerve cell that represents the main attribute of sweetness is connected to a nerve cell that represents the subattribute of the degree of sweetness.
- This sub-attribute may be connected to a main attribute other than sweetness to modify the main attribute. Attributes modified from other attributes are main attributes, and attributes that modify other attributes are sub-attributes.
- Attribute may be composed of only one main attribute or one main attribute and one or more sub-attributes. This corresponds to the case of modification with an adjective and an adverb.
- the feature data of sweetness sends a signal to a nerve cell representing the attribute of sweetness.
- the characteristic data of the degree of sweetness sends a signal to a nerve cell that represents an attribute associated with the degree of sweetness.
- the degree of sweetness can be understood by imagining the taste of an aqueous sugar solution. The higher the sugar concentration, the greater the degree of sweetness. The lower the sugar concentration, the less sweetness.
- a nerve cell that represents an attribute associated with the sweetness that received the signal and a nerve cell that represents the attribute associated with the degree of sweetness received the signal are connected by a dendrite or the like. In this way, an object can have an attribute and a degree of the attribute, that is, a main attribute and a sub attribute.
- taste information is received from the gustatory nerve, and feature data is created in which information representing the strength of each taste type is arranged in a certain order, such as bitter after sweetness and sour after bitterness.
- the feature data is connected to specific nerve cells in the attribute region 115 in the order of the taste feature data.
- specific nerve cells in the attribute region 115 express a specific taste.
- Characteristic data of individual tastes such as sweetness and sourness of a certain object can also be obtained. Attributes that express individual tastes can be used by all things in common. Further, if the pattern of the entire feature data of tastes arranged in a certain order is converted into one feature data and connected to one attribute, the attribute of the entire taste of the object can be created.
- a nerve cell associated with an attribute is referred to as an attribute nerve cell
- a nerve cell associated with an object is referred to as an object nerve cell.
- a neuron associated with the main attribute is called a main attribute neuron
- a neuron associated with the sub attribute is called a sub attribute neuron.
- the information is received from the sensory organ 110 which is an information input unit for inputting information, the feature extraction unit 114 extracts the feature from the received information, and the extracted feature is compared with the feature data already stored in the storage area 112. .
- information such as sound is input from the sensory organ 110, and a plurality of features such as sound pitch, sound volume, frequency distribution, and sound waveform are extracted by the feature extraction unit 114. If there is feature data that is the same as or similar to the extracted feature, a signal is sent from the feature data to the attribute connected to the feature data. This process is repeated for the newly generated feature.
- the one or more attributes that receive the signal send a signal to the object connected to the attribute.
- One object that has received the signal is recognized as an object that has provided information to the sensory organ 110.
- the object that has received the most signals is recognized as the object that has given information to the sensory organ 110.
- This procedure is a procedure for performing recognition.
- a collection of things with the same attributes is a collection of knowledge.
- the relationship between things is expressed by connecting things.
- a connected knowledge network is a knowledge network. The meaning of the relationship varies depending on the knowledge network. For example, the relationship between objects can express various relationships such as parent-child relationships, time context, and inclusion relationships.
- a knowledge network may be created by connecting things other than the knowledge collection to the knowledge network in the knowledge collection.
- a knowledge network may be created by selecting things from all things without being limited to those in a certain knowledge collection, and connecting the selected things.
- Knowledge network is also an abstract object. Therefore, it is possible to generate a knowledge collection or knowledge network including a knowledge network as an abstract object.
- a nerve cell 70 associated with an object in the object region 111 is also called an object.
- the nerve cell 71 in the attribute region 115 is called an attribute.
- the object area 111 includes all objects. In this state, it is difficult to find a specific object from the object region 111. You can choose several things as follows. Send a signal from an attribute to the object connected to that attribute. An object is activated when it receives a signal from an attribute. The activated object has the attribute. In this way, a collection of objects having certain attributes can be generated. This is called a knowledge collection.
- the attribute for sending a signal is not limited to one, and may be one or more.
- Attribute (4) is an attribute called the Suzuki family.
- the object (F) represents the father of the Suzuki family
- the object (G) represents the mother of the Suzuki family
- the object (H) represents the child of the Suzuki family.
- a thing (H) that is a child of the Suzuki family is connected from a thing (F) that is the father of the Suzuki family, and a thing (H) that is a child of the Suzuki family from the thing (F) that is a mother of the Suzuki family.
- connection may be bidirectional. However, many relationships are thought to have direction. Therefore, a single direction, that is, a connection with a direction is common. Suppose two things are connected in a single direction. When this unidirectional connection is associated with the inclusion relationship, for example, the connection starting point object includes a thing (superordinate concept), and the connection arrival point object includes a subordinate concept. Of course, it can be reversed.
- a knowledge network with the same family can create various knowledge networks, such as a knowledge network linked by height, a knowledge network linked by age, and a knowledge network linked by parent-child relationship.
- a knowledge network connected by a height order relationship is characterized by an attribute of height in addition to the attribute of XX house.
- Knowledge network may express the relationship of components such as “water is made up of hydrogen and oxygen”.
- the knowledge network may represent an inclusive relationship such as “human beings are included in mammals”. It is assumed that the object A and the object B are connected only from the object A to the object B and represent an inclusion relationship. For example, the object A is a mammal and the object B is a human.
- the thing A has an attribute common to mammals and an attribute representing mammals, and the common attribute possessed only by humans, the attribute representing mammals in the superordinate concept, and the attribute representing humans that are the concepts of the thing B itself are subordinate concepts. Make thing B have. In this way, the thing representing the lower concept may not have the attributes of the higher concept.
- the attribute which the attribute which the thing which represents the superordinate concept has, and the attribute which the thing B which represents the subordinate concept has has the attribute which the thing which represents the subordinate concept has. In this way, the connection between objects and attributes can be reduced. What is necessary is just to acquire the attribute which the thing showing the high-order concept has as needed.
- One-way connections can also be related to blood relationships. These are examples, and various relationships can be expressed by the connection of things in one direction.
- the knowledge collection or knowledge network it is possible to check whether the information input from the outside is consistent with the knowledge collection or knowledge network knowledge. For example, assume that one object and several attributes are input from the outside. Generate a knowledge collection with all the attributes entered. If the given object is included in the knowledge collection, it can be understood that the object has the input attribute. In addition, if the following is performed, it is possible to determine whether there is a relationship between two given objects. If one thing can trace the connection of the knowledge network and reach the other, it can be seen that there is a relationship between the two. Thus, the relationship between the two objects given can be acquired.
- a signal is sent from the attribute of the family of the Suzuki family. Everything that receives a signal has the attribute of the Suzuki family. The whole thing that receives the signal becomes a collection of knowledge with the attribute of the Suzuki family. If there is a thing called Hanako in the generated knowledge collection, Hanako can be determined to be included in the Suzuki family. If there is no such thing as Hanako in the generated knowledge collection, it can be determined that Hanako is not included in the Suzuki family.
- the left frame in FIG. 45 is a knowledge network representing the Yuzo Suzuki family.
- the right frame in FIG. 45 is a knowledge network representing the Jiro Suzuki family. Mr. Jiro belongs to these two knowledge networks.
- the Yuzo, Yue, Ichiro, and Jiro members of the Yusan family are connected to the attributes representing the Yuzo family.
- the Jiro, Jika, Mika, Taro, and Hanako of the Jiro Suzuki family are connected to the attributes that represent the Jiro family of Suzuki.
- the two knowledge networks have a common thing called Jiro.
- the attribute representing the Yuzo family of Suzuki is connected to feature data generated from a video in which, for example, Yuzo, Megumi, Ichiro, and Jiro are together. Further, the object representing Yuzo is connected to, for example, an attribute connected to feature data generated from Yuzo's video.
- a word representing the name “Yuzo” is described, but this is described only for explanation.
- Jiro also belongs to a knowledge network called Yuzo Suzuki. Therefore, it moves from a knowledge network called Jiro Suzuki to a knowledge network called Yuzo Suzuki through a thing called Jiro. Then, in the knowledge network called Yuzo Suzuki, Jiro moves to Yuzo. In this way, since it is possible to move from Hanako to Yuzo of another knowledge network, there may be some relationship between Hanako and Yuzo. In fact, Yuzo is Hanako's grandfather. Thus, inference can be made using not only one knowledge network but also connected knowledge networks.
- the use of knowledge collection is effective for finding relationships between things. There is a possibility that a collection of knowledge having a common thing is related.
- the use of a knowledge network is also an effective method for finding out whether there is a relationship between things. If a route is found between two things, then the attributes of the two things are compared to see if the two things have the same or similar attributes. If there is the same attribute, it can be determined that two things are related to the attribute. If there is a similar attribute, it can be determined that there is a high possibility that the two objects are related. Thus, in this embodiment, inference can be performed without identification data generated from words.
- ⁇ ⁇ ⁇ Use the specified object as a starting point to find an object that can be moved from the starting point.
- the attributes of the found object and the specified object that is the starting point are compared to see what common or similar attributes the two objects have. If a common attribute is found, it can be seen that two things are related by a common attribute. If similar attributes are found, it can be seen that two things may be related by similar attributes. However, just because there are many common or similar attributes does not mean that there is an important meaningful relationship. Only one common attribute or only one similar attribute has an important meaning, and a connection between a newly found object and an object may be an important and meaningful relationship.
- a collection of things with common attributes is a collection of knowledge.
- a signal is sent from a neuron of a certain attribute to a neuron of an object in the domain of the object. All the nerve cells of the object that receives the transmitted signal constitute an object of knowledge about the attribute.
- the thing with which the attribute is associated is knowledge.
- an attribute nerve cell may be referred to as an attribute
- an object nerve cell may be referred to as an object.
- Specified attribute to generate a collection of knowledge about the attribute. Further, another attribute is specified for the generated knowledge collection, and a new knowledge collection is generated by collecting objects having the attribute. By repeating this, a collection of knowledge can be generated by specifying a plurality of attributes.
- This connected collection of things is a knowledge network.
- This connection can be meaningful. For example, inclusive relationships such as “human being a mammal”, blood relationships such as “Yuzo is Jiro's father”, positional relationships, time relationships, etc. Can be expressed.
- FIG. 16 shows an example of generating a knowledge collection related to a certain attribute.
- the object (A) belongs to both the knowledge collection generated by the attribute (1) and the knowledge collection generated by the attribute (2). In this way, a collection of knowledge can be created for each of the various attributes of the object, and the knowledge can be organized and used.
- Making things and attributes, connecting things, attributes, and connecting things and attributes are learning. Learning is also creating feature data, connecting feature data and objects, and feature data and attributes. Learning is also creating identification data, connecting identification data and objects, and identification data and attributes. It is also learning to delete the connection between objects, attributes, feature data, and identification data.
- a feature is extracted by the feature extraction unit 1020 from the input information. If there is feature data of the same feature in the storage area and the feature data is connected to the object via the attribute of the attribute area, a signal is sent from the feature data to the object. When there is feature data similar to the extracted feature, a signal is sent in the same manner, and the object receives the signal. The one that receives the signal is activated. This is a method for recognizing an object. Signals reach several objects from multiple feature data via attributes. The thing that receives the strongest signal is the first candidate for that thing.
- the feature extraction unit in FIG. 5 extracts features from the sound received by the ear which is a sensory organ.
- the data comparison unit in the brain corresponding to the data comparison unit 1030 compares the extracted features with the feature data held in the storage area. Then, the data comparison unit in the brain sends a signal to the attribute corresponding to the matching or similar feature data, and the attribute receiving the signal sends the signal to the object, so that the object is recognized by the sound heard by the ear. .
- a knowledge collection is generated by specifying an attribute identifier of one or more attributes given in advance.
- the specified attribute is a condition for an object to be included in this knowledge collection.
- the knowledge collection is searched for whether or not the given object is included in the knowledge collection. If it is included, it can be determined that the given object satisfies the condition.
- Inference can be made using the knowledge network of this embodiment.
- Objects in this knowledge collection have the attribute of illness.
- the attribute of illness is connected to characteristic data generated from images of people who are sleeping due to illness and characteristic data of various symptoms when ill.
- the abstract attribute of illness is connected with the attribute of illness.
- the feature data is generated, and it is checked whether the same feature data is in the knowledge base. If there is the same feature data, the attribute of illness connected to the feature data of the face having a poor facial color is acquired. Next, an abstract object called “disease” connected to the attribute of “disease” is acquired and recognized as a disease.
- This knowledge collection includes individual diseases such as colds and pneumonia.
- Each disease such as a cold and tuberculosis has a disease name and symptoms as attributes.
- the attribute of a cold disease name is connected to identification data generated from a video image of the word “cold” and identification data generated from the sound of the word “cold”.
- the identification data representing the above-mentioned disease name itself is not used as a meaning or concept relating to the disease.
- a cold has an attribute of swelling of the tonsils.
- the attribute of swelling of tonsils is connected with feature data of swollen tonsils video.
- the concept and meaning are generated by connecting the feature data to the attribute.
- Feature data generated from images of many types of tonsils can be linked to one attribute such as swelling of the tonsils. The more types of feature data are connected to one attribute, the more appropriate the judgment can be made.
- an image of swollen tonsils is input from the outside to the knowledge base system, features are extracted from the images, and feature data is generated. Next, it is searched whether there is feature data determined to be the same as the generated feature data in the knowledge base. When feature data determined to be the same is found, an attribute connected to the feature data is acquired.
- a diagnosis education system and a diagnosis support system can be created by displaying some of the acquired diseases connected with the attribute on the display device.
- a knowledge network can be created in this knowledge collection. It is possible to generate a knowledge network where various diseases are systematically connected starting from the disease. A knowledge network that links related diseases can also be created. For example, in a knowledge network, a cold and pneumonia are connected. It expresses the relationship between the two diseases that pneumonia occurs when a cold is squeezed in a one-way connection. By using such a knowledge network, it is possible to display a related disease name and support diagnosis. For example, a patient with a cold symptom can be advised to check with a doctor for signs of pneumonia.
- the technology of the present application can be used to construct a judgment work support system or a judgment work education support system in a work to make any judgment.
- voice recognition There is a technology called voice recognition.
- This technology can also be used for voice recognition.
- Features are extracted from speech, the extracted features are stored as identification data, and the identification data is associated with objects and attributes in advance.
- Objects, attributes, and identification data are connected in the above-described three-layer structure. In this way, based on the input voice, not only the identification data of the word represented by the voice but also the concept, meaning and object corresponding to the voice can be acquired.
- Features are extracted from the input speech and identification data is generated. It is possible to acquire and recognize attributes and objects connected to identification data that is similar or coincident with the generated identification data. Other attributes connected to the recognized object can be acquired, and feature data and identification data connected to the attribute can be acquired.
- a speech recognition system can be implemented by extracting features from the video. If features are extracted from sensors such as a temperature sensor and a pressure sensor, they can also be used as a facility monitoring system. Of course, it can also be used for information written in text.
- FIG. 46 shows a triangle 121 as an example of a figure. It may be a figure other than a triangle. Moreover, not only a plane figure but a solid figure may be sufficient.
- the triangle 121 includes three lines (straight lines) 123, three points (vertices) 122 at which the three lines 123 intersect with each other, and a plane 124 surrounded by the three lines 123. Instead of the line 123, a figure can be formed by a curve.
- the surface 124 may be a curved surface instead of a flat surface. Considering the line 123, the point 122, and the surface 124 as objects, a knowledge network composed of the line 123, the point 122, or the surface 124 is created.
- FIG. 47 is a knowledge network representing a triangle 121 by three points 122. Since each point 122 is a thing, it can have an attribute. The point 122 can have, for example, a coordinate value as an attribute.
- FIG. 48 is a knowledge network representing a plane figure of a triangle 121 with points 122 and lines 123.
- a figure is represented by a knowledge network in which points 122 and lines 123 are alternately connected.
- the point 122 has a coordinate position and an interior angle as attributes.
- the line 123 has the length of the line as an attribute.
- the attribute is not limited to the coordinate position or length.
- FIG. 49 shows a knowledge network in which a plane figure called a triangle 121 is connected to a line 123 that is a component of the figure. All the components of the graphic may be connected to the plane, or one or more components may be connected to the plane. Of course, it may contain things called dots.
- ⁇ As attributes of the plane it may have a feature that the area, the length of two sides are equal, the name of an isosceles triangle, and the like.
- a graphic element is connected to feature data generated from a video to generate its meaning and concept.
- An attribute which is a feature that the lengths of two sides are equal leads to feature data generated from an isosceles triangle image.
- the attribute representing the name of an isosceles triangle is connected to identification data generated from a video image of the word isosceles triangle and identification data generated from a sound of an isosceles triangle.
- the attribute representing only the identification data generated from the word isosceles triangle is not connected to the feature data representing the meaning or concept of the isosceles triangle.
- an attribute that has no meaning and is only for identification is clearly separated from an attribute that represents the concept or meaning of an object.
- FIG. 50 is a three-dimensional figure called a triangular pyramid, and is composed of four triangles (surfaces) 121.
- FIG. 51 is an example in which a solid is represented by a knowledge network, taking a triangular pyramid as an example.
- a triangle (surface) 121 that is a three-dimensional component is connected to a graphic element constituting the surface.
- the lines (sides) 123 shared by the respective faces are connected to each other. In this way, it is possible to represent how the surfaces constituting the solid are in contact.
- the same points and lines can be used on a plurality of surfaces for the shared components such as points and lines.
- Several representation methods are possible, such as a three-dimensional development view.
- FIG. 52 shows the eyes 126, the nose 127, and the mouth 128.
- FIG. 53 shows an example of a knowledge network in which a straight line connecting the eyes, a straight line connecting the eyes and the nose, and a straight line connecting the nose and the mouth are connected to each other.
- a knowledge network that represents the graphical features of the face.
- Each line has a line length as an attribute. From this knowledge network, ratios such as the length between eyes and the length from eyes to nose can be obtained. These ratios become facial features.
- features are extracted from the image of the face given from the outside, a knowledge network as shown in FIG. 53 is generated, and the distance ratio between eyes, nose and mouth is obtained. The same ratio is calculated for the face knowledge network held in advance, and the two are compared. In this way, it is possible to determine whether or not the face image given from the outside is similar to the stored face.
- FIG. 54 is a knowledge network representing a shogi board.
- a place 129 where a shogi piece is placed is a thing. This is a knowledge network where the places 129 are connected like a grid.
- the place 129 where the piece is placed has information indicating the position on the board as an attribute. For example, it is represented by the vertical and horizontal position from one corner of the board as (1, 1).
- the type of the placed piece is added to the attribute of the place 129 where the piece is placed.
- King, Ayumu, etc. are the types of pieces.
- the state where there is no piece is expressed by having an attribute indicating that it is vacant or having no piece attribute. When a piece moves, the attribute of that piece is moved to the destination.
- the original place should have no pieces.
- As an attribute of the place 129 where the piece is placed various information such as a value indicating the degree of safety and importance of the place can be held.
- a thing that represents a piece outside the figure is a thing, and has attributes such as the kind of piece and the movable range.
- a place 129 where a piece at a position where each piece can move may have an attribute such as “movable walk”.
- the place 129 where the piece that is an object is placed may be connected to the piece that is an object, and the piece itself may have attributes such as the movement operation range of the piece. A more suitable method can be adopted according to the purpose.
- Judgment or reasoning can be made based on the separation of objects.
- Judgment or inference can be made based on the position in the space created by connecting objects. This corresponds to a big picture.
- an abstract object representing the meaning or concept represented by the attribute can also be handled.
- the attribute connected to the feature data representing the meaning of the abstract object is connected to the object.
- an abstract object called “heat” an attribute is connected to characteristic data of the sense of heat when touching a hot object, and that attribute is connected to the object.
- the abstract object “heat” can be connected to the identification data generated from the word “heat”, but may not have the attribute connected to the identification data generated from the word “heat”.
- identification or selection can be performed using the identification data generated from the word.
- An object directly connected to the identification data can be identified by the identification data.
- the attribute connected to the identification data can be identified by the identification data.
- the object can be indirectly selected and identified by the identification data.
- the knowledge base system can construct a knowledge base from feature data, attributes, and objects without identification data, accumulate knowledge, and perform knowledge processing such as inference.
- a knowledge base is constructed from identification data, feature data, attributes, and objects, knowledge is accumulated, and knowledge processing such as inference can be performed. It is also possible to construct a knowledge base consisting of identification data, attributes, and objects without feature data, accumulate knowledge, and perform knowledge processing such as inference. In this case, however, the knowledge base has no meaning or concept.
- a knowledge base according to an embodiment of the present technology includes identification data or feature data, attributes, and objects.
- the relational database 14 can handle tables efficiently. However, other software that can handle connections may be used. Software that can handle a data structure called a list structure is also possible, but in order to realize the list structure, data in the form of a table is used.
- a list structure is represented by a connection pointer between elements as used in a data structure that realizes a node that is an object. When representing a connection in a table, a table in which connected elements are arranged in one row is created. The difference is whether the connection is expressed by the pointer or the elements are arranged in one line to express the connection.
- the associative memory may be a device such as an associative memory or an electronic component such as an integrated circuit.
- the associative memory has a table inside and performs the following processing by hardware. Data is received from outside the associative memory, and it is checked whether there is data in the table that matches part or all of the received data. If there is matching data in the table, data associated with the matching data is acquired from the table in the associative memory, and the acquired data is output.
- a table in the associative memory a table representing connections such as a table in which objects are associated with each other, a table in which objects are associated with attributes, and a table in which attributes are associated with feature data is created.
- the associative memory is an electronic component such as a device or an integrated circuit having a search, addition, and deletion function of the relational database 14 and tabular data.
- the network in FIG. 55 is an embodiment, and is not limited to the network in FIG.
- Such a network is called a search network.
- identifiers can be associated using a network as shown in FIG. 55 as shown in FIG. 55 . You can use this network to search, add, and delete associated identifiers.
- the network in FIG. 55 is one of the auxiliary storage devices 7 in FIG. 7 that holds only the table.
- the data equivalent to the table in the relational database 14 is decomposed and held in the communication device in the network in FIG. A search process equivalent to 14 is performed. Searches can be performed faster than the relational database 14.
- a normal network transmits information including a communication device identifier that is a destination, and a communication device whose destination matches the communication device identifier receives the information.
- information including the communication device identifier of the transmission source is transmitted without including the destination.
- data can be transmitted to a partner designated in advance without designating a recipient.
- FIG. 55 shows the network 151.
- the network 151 in FIG. 55 includes a transmission device 152, a communication device 156, a control unit 160, and a transmission reception unit 161 connected to each other via a communication path 155.
- the transmission device 152 includes a transmission unit 153.
- the communication device 156 includes a reception unit 157, a communication device identifier 158, a receivable identifier 159, a search unit 162, and an address 163.
- this network 151 information is transmitted to the communication device 156 using the shared transmission device 152.
- the transmission device 152 and the transmission unit 153 may be mounted so as to be included in each communication device 156. However, since only the transmission unit 153 of one transmission device 152 can transmit at the same time, it is desirable to share one transmission device 152 and transmission unit 153 with a plurality of communication devices 156.
- the reason for limiting the number of transmission devices 152 to one is that transmission data does not collide on the communication path 155 and transmission efficiency is greatly improved. Since transmission is simultaneously performed from the transmission device 152 to the plurality of communication devices 156, the efficiency is further improved.
- the transmission reception unit 161 receives a transmission instruction and transmission data from the control unit 160.
- the transmission data does not include a destination, and includes only one communication device identifier 158 that is an identifier of the transmission source communication device 156.
- One communication device 156 is associated with one object, attribute, feature data, and identification data.
- the transmission unit 153 includes only one transmission source communication device identifier 158 received by the transmission reception unit 161 and transmits information including no destination to the communication path 155.
- the communication path 155 connects the transmission device 152 and the communication device 156.
- the communication path 155 may be a communication path 155 in addition to electric wires and optical cables.
- the communication path 155 may be anything.
- the communication device 156 includes a reception unit 157, a communication device identifier 158 that is an identifier of the communication device 156 itself, a receivable identifier 159 that is an identifier of the receivable communication device 156, and a search unit 162.
- the communication device identifier 158 is an identifier for identifying the communication device 156 in one network, and is set in the communication device 156 by the control unit 160 in advance.
- the receivable identifier 159 is a communication device identifier 158 of another communication device 156 corresponding to another nerve cell connected to the communication device 156 corresponding to one nerve cell.
- the receivable identifier 159 is preset in the communication device 156 by the control unit 160.
- the control unit 160 can additionally delete the receivable identifier 159 of the communication device 156.
- the communication devices 156 corresponding to nerve cells can be connected to each other, or the connection can be deleted.
- the receiving unit 157 receives the communication device identifier 158 of the transmission source in the data transmitted from the transmission device 152 from the communication path 155.
- the search unit 162 checks whether the received communication device identifier 158 matches any of the receivable identifiers 159. If the received communication device identifier 158 matches with any of the receivable identifiers 159, the search unit 162 outputs data including the communication device identifier 158 of the communication device 156 itself to the control unit 160. This corresponds to the communication device 156 corresponding to the nerve cell receiving a signal from another communication device 156 that is another connected nerve cell.
- the communication device 156 does not output the communication device identifier 158. In this case, it represents that the communication device 156 corresponding to the nerve cell identified by the received communication device identifier 158 is not connected to the communication device 156 corresponding to the nerve cell.
- a table in which a certain communication device identifier is associated with another communication device identifier is searched with a certain communication device identifier, and another communication device identifier associated with the communication device identifier is obtained. The same processing can be performed. In other words, the table can be searched using this network.
- control unit 160 When the control unit 160 receives the communication device identifier 158 output from the search unit 162, the data including the received communication device identifier 158 is passed to a device outside the figure.
- the control unit 160 receives a transmission instruction including a transmission source communication device identifier 158 that is an identifier of a communication device corresponding to a nerve cell from an unshown device, and sends the received transmission instruction to the transmission reception unit 161.
- the address 163 of each communication device 156 uniquely identifies the communication device 156.
- the control unit 160 designates the communication device 156 with the address 163 and sets data including the communication device identifier 158 and the receivable identifier 159.
- the computer or knowledge base system outside the figure sends data including the communication device identifier 158, the receivable identifier 159, and the address 163 of the communication device 156 to the control unit 160, and sends the communication device identifier to the communication device 156 specified by the address 163.
- 158, data including a receivable identifier 159 is set.
- the communication device identifier 158 identifies the communication device 156 within the network 151, and the address 163 of the communication device 156 is used to identify the communication device 156 from outside the network 151.
- a transmission instruction including the communication device identifier 158 of the transmission source is sent from the device outside the figure to the controller 160 device.
- the network 151 performs the above-described processing.
- the control unit 160 Upon receiving the communication device identifier 158 of another communication device 156 connected to the communication device 156 identified by the communication device identifier 158 of the transmission source included in the transmission instruction, the control unit 160 communicates with the communication device of the other communication device 156 connected. Data including the identifier 158 is sent to a device not shown.
- the search processing in each communication device 156 can be performed in parallel, and the search processing for each communication device 156 is performed in hardware, so that the table can be searched very quickly. Yes. For this reason, the processing speed is remarkably increased. It can also be used in fields where a quick response is required.
- the communication device 156 of this network has a simple structure and can be easily implemented as an integrated circuit. It is also easy to incorporate a plurality of communication devices 156 in one integrated circuit.
- a network is generated by connecting a plurality of integrated circuits having a plurality of communication devices 156 in parallel, a large table can be created.
- a knowledge base system that can store large amounts of knowledge information and process it at high speed can be constructed.
- it is necessary to periodically perform maintenance work for maintaining the performance of the database, but this network 151 does not have to do so.
- control unit 160 7 and the network 151 of FIG. 55 are connected by the control unit 160, and instructions and data are passed from the calculation device 2 to the control unit 160 of FIG.
- the arithmetic device 2 receives data from the control unit 160.
- the control unit 160 receives the received communication from the communication device 156 at the address 163. Data including the device identifier 158 and the receivable identifier 159 is set.
- the control unit 160 transmits the transmission instruction and the transmission source communication device identifier 158 to the transmission reception unit 161.
- the transmission accepting unit 161 sends a transmission instruction and a transmission source communication device identifier 158 to the transmission unit 153 of the transmission device 152.
- the transmission unit 153 transmits the source communication device identifier 158 to the communication path 155.
- the communication device 156 compares the received communication device identifier 158 with the receivable identifier 159.
- the communication device identifier 158 of the communication device 156 is passed to the control unit 160.
- the communication device identifier 158 of the communication device 156 and the data directly associated with the receivable identifier 159 are obtained. To the control unit 160.
- the control unit 160 passes the data including the passed communication device identifier 158 to the arithmetic device 2.
- the communication device 156 passes the data including the communication device identifier 158 of the communication device 156 and the received communication device identifier 158 of the transmission source to the control unit 160.
- the control unit 160 may pass a set of data including the communication device identifier 158 of the communication device 156 and the communication device identifier 158 of the transmission source to the arithmetic device 2.
- the communication device identifier 158 is an identifier of an object, attribute, feature data, or identification data. That is, one communication device 156 is one thing, attribute, feature data, or identification data. However, the actual data of the feature data is stored in the data 10 other than the network 151 in FIG. 55 together with the identifier of the feature data, and is referred to by the identifier of the feature data. The actual data of the identification data is also stored in the data 10 other than the network 151 in FIG. 55 together with the identifier of the identification data, and is referred to by the identifier of the identification data.
- the receivable identifier 159 is, for example, an identifier of another object, other attribute, other feature data, or other identification data connected to one object that is the communication device 156.
- the communication device 156 represents an object, an attribute, feature data, or identification data.
- the network 151 represents a connection between things, attributes, feature data, and identification data. Data set in the communication device 156 is held as a table of the relational database 14 or the like.
- connection table For the connection table, select one of the associated columns, select one identifier of the selected column, and extract the identifier of the other column that is connected and associated with that identifier. A new table is created that associates only one identifier with some extracted identifiers. This process is repeated for the column identifier. If each row of the connection table is associated with additional information such as a weight indicating the degree of connection, a table is created by further associating the additional information with the other column identifier. Also good.
- the attached information is not limited to the weight.
- the latter column which is the other column
- one identifier of the selected column is selected
- the identifier of the former column which is one column linked and associated with the identifier
- a new table is created that associates only one identifier with some extracted identifiers. This process is repeated for the column identifier. If each row of the connection table is further associated with attached information such as a weight indicating the degree of connection, a table is created by further associating the attached information with the associated identifier of the former column. Also good.
- the attached information is not limited to the weight.
- a newly created table corresponds to one object, attribute, feature data, and identification data.
- This table associates, for example, one object identifier with an identifier of another object, attribute, feature data, or identification data connected to the one object. Therefore, a combination of the identifier of the one thing, attribute, feature data or identification data and a table having one column holding the identifier of the thing, attribute, feature data or identification data may be used.
- the number of identifiers representing other connected objects, attributes, feature data, or identification data is 0 or more and is not limited to one.
- FIG. 56 is a table associating one feature data with several attributes in FIG. 57 and a table associating one attribute with several feature data in FIG. Can be disassembled.
- the table stores attribute identifiers instead of attributes, and feature data identifiers instead of feature data.
- One feature data identifier (10001) is acquired from the feature data identifier column, which is one column of the table associating the feature data with the attributes in FIG.
- the attribute identifier (14) associated with the feature data identifier (10001) is extracted and acquired from the attribute identifier column which is the other column.
- a table is generated in which only one identifier (10001) of the acquired feature data is associated with the extracted attribute identifier (14). This is the table of FIG. If each row in the table of FIG. 56 has attached information such as a weight and is associated with each row, the associated information is further associated with the attribute identifier (14) associated with the feature data identifier (10001). Then, the table of FIG. 57 may be created. The above processing is repeated for all feature data identifiers in the feature data identifier column, which is one column of the table associating the feature data with the attributes in FIG.
- One attribute identifier (14) is acquired from the attribute identifier column in the table associating the feature data and attributes shown in FIG.
- the feature data identifier (10001) associated with the attribute identifier (14) is extracted and acquired from the feature data identifier column. Further, the identifier (45000) of the feature data associated with the attribute identifier (14) is extracted and acquired.
- a table is generated in which only one acquired attribute identifier (14) is associated with the extracted and acquired feature data identifiers (10001, 45000). This is the table of FIG. If each row in the table of FIG. 56 has attached information such as a weight and is associated with each row, the attached information is further associated with the feature data identifier (10001) associated with the attribute identifier (14). 58 may be created. The above process is repeated for all the attribute identifiers in the latter attribute column of the table associating the feature data with the attributes in FIG.
- one table associating the feature data with the attributes in FIG. 56 includes one feature data, a table representing several attributes directly associated with the one feature data, and one table. It is broken down into a table that represents an attribute and some feature data directly associated with that one attribute.
- One of these decomposed tables is assigned to one communication device 156.
- the identifier of one feature data is set as the communication device identifier 158
- several attribute identifiers are set as the receivable identifier 159.
- the attached information is directly associated with the attribute identifier
- the attached information is directly associated with the attribute identifier, that is, the receivable identifier 159, and set in the communication device 156.
- the table of the receivable identifier 159 is not a one-column table but a multi-column table, and the attached information is set in the same row as the associated attribute identifier.
- the data 10 includes a list of unused identifiers that hold unused object identifiers, unused attribute identifiers, unused feature data identifiers, and unused identification data identifiers.
- the object identifiers and attribute identifiers used in the table associating the objects with attributes in FIG. 19 are obtained from the unused identifier list.
- Both the identifier of the feature data and the identifier of the identification data are obtained from the unused identifier list.
- the communication device 156 and the address 163 of the communication device are associated with each other in a one-to-one correspondence in advance, and only one communication device 156 can be designated by the address 163 of one communication device.
- the data 10 includes a list of unused communication devices that hold addresses 163 of unused communication devices. Further, the data 10 includes an address identifier correspondence table in which the communication device address 163 is associated with an object identifier, an attribute identifier, a feature data identifier, or an identification data identifier.
- the address 163 of the communication device is acquired from the identifier of the object, attribute, feature data, or identification data using the address identifier correspondence table, and the acquired address Data including 163 and an instruction are passed to the control unit 160.
- an identifier is acquired from the unused identifier list
- an address 163 of the communication device is acquired from the unused communication device list
- a pair of the acquired identifier and address 163 is obtained. Stored in the address identifier correspondence table.
- one communication device 156 corresponds to one thing, one attribute, one feature data, or one identification data. Note that the actual data of the feature data or the identification data, which is the feature extracted from the voice or video, is held in the data 10 other than the network 151 in FIG.
- the arithmetic device 2 passes the data including the address 163 and the communication device identifier 158 set to the communication device 156 specified by the address 163 and the receivable identifier 159 to the control unit 160 of the network 151, and is specified by the address 163.
- the communication device 156 is instructed to set data including the communication device identifier 158 and the receivable identifier 159.
- control unit 160 Upon receiving the instruction, the control unit 160 sets data including the communication apparatus identifier 158 and the receivable identifier 159 in the communication apparatus 156 specified by the address 163 according to the received data and instruction.
- the communication device address 163 and the communication device identifier 158 are associated with each other and stored in the address identifier correspondence table.
- the table associating the feature data with the attributes in FIG. 56 is a table associating one feature data with several attributes in FIG. 57 and a table associating one attribute with several feature data in FIG. Can be disassembled.
- the table stores attribute identifiers instead of attributes, and feature data identifiers instead of feature data.
- One unused address 163 is acquired from the unused communication device list.
- a set of the acquired address 163 and the feature data identifier (10001) in the first table of FIG. 57 is stored in the address identifier correspondence table.
- the address 163 is associated with the identifier of the feature data. If the address 163 is already associated with the feature data identifier (10001), the address 163 associated with the feature data identifier (10001) is acquired from the address identifier correspondence table.
- the arithmetic device 2 sets the acquired address 163, the feature data identifier (10001) as the communication device identifier 158, the attribute identifier (14) as the receivable identifier 159, the weight (1) as the attached information, and the setting
- the instruction is passed to the control unit 160.
- the control unit 160 sets the identifier (10001) of the feature data received as the communication device identifier 158 for the communication device 156 specified by the passed address 163, and receives the attribute identifier (14) received as the receivable identifier 159.
- a weight (1) is set as ancillary data of the attribute identifier (14).
- the setting is repeated in the same manner for all the tables in FIG.
- the table associating the feature data and the attributes in FIG. 56 is stored in the network in FIG.
- One selected network is a table associating identifiers and representing connections as shown in FIG.
- the control unit 160 passes the received transmission source identifier and transmission instruction to the transmission reception unit 161.
- the transmission reception unit 161 passes the received transmission source identifier and transmission instruction to the transmission unit 153 of the transmission device 152.
- the transmission unit 153 transmits the received transmission source identifier (14) to the communication path 155 in accordance with the received transmission instruction.
- the transmission source identifier (14) is received from the communication path 155.
- the search unit 162 checks whether the received transmission source identifier (14) matches any of the receivable identifiers 159. In this case, the received transmission source identifier (14) matches one of the set receivable identifiers 159.
- the feature data identifier (10001) which is the communication device identifier 158 of the communication device 156 and the weight (1) which is the attached data are passed from the communication device 156 to the control unit 160.
- the search unit 162 checks whether the received transmission source identifier (14) matches any of the receivable identifiers 159. In this case, the received transmission source identifier (14) matches one of the set receivable identifiers 159.
- the communication device 156 passes the feature data identifier (45000), which is the communication device identifier of the communication device 156, and the weight (2), which is the attached data, to the control unit 160.
- the computing device 2 includes a set of feature data identifier (10001) that is data passed from the communication device 156 to the control unit 160 and weight (1) that is attached data, a feature data identifier (45000), and attached data. A pair with a weight (2) is received. It can be seen that the feature data (10001) and the feature data (45000) are linked to the attribute identifier (14). Furthermore, the weight representing the strength of connection can also be acquired as attached information of the feature data.
- the above processing is the same as the table in FIG. 56 being searched using the relational database 14 and using the attribute identifier (14) as a search key.
- the communication device 156 may not send the attached data to the control unit 160, or special data indicating that there is no attached information, For example, NULL may be passed to the control unit 160.
- a program represents a processing procedure with information called a program. It is possible to make this hardware. Accordingly, the program of the present application can be implemented by hardware. By implementing the program of the present technology as an electronic device, an integrated circuit, or a processor implemented by hardware, the present technology can be incorporated into various devices and can be processed at high speed. Of course, it can be implemented as an information system.
- the present invention is advantageously used in a knowledge base system.
Abstract
Description
(1)山の、座標軸1は1
(2)岩塩の、座標軸2は1、座標軸3は-1
(3)岩塩でできた山の、座標軸1は1、座標軸2は1、座標軸3は-1
同じ内容を、記号を用いて次のように記述することもできる。
(1)山(1:1)
(2)岩塩(2:1、3:-1)
(3)岩塩でできた山(1:1、2:1、3:-1)
(1)山(1)
(2)岩塩(2)
(3)岩塩でできた山(1、2)
(1)山(1:1)
(2)岩塩(2*9:1、3:-1)
(3)岩塩でできた山(1:1、2*9:1、3:-1)
(1)A(1:1)
(2)B(2*9:1、3:-1)
(3)C(1:1、2*9:1、3:-1)
Aの代わりに1000、Bの代わりに1001、Cの代わりに1002を使うと次のようになる。
(1)1000(1:1)
(2)1001(2*9:1、3:-1)
(3)1002(1:1、2*9:1、3:-1)
さらに、真偽値が真の属性だけを記述すると次のようになる。
(1)1000(1)
(2)1001(2*9)
(3)1002(1、2*9)
B.include((2*9:1))
B.isLike((2*9:1))
2 演算装置
3,156 通信装置
4,151 ネットワーク
5 外部記憶装置
6 主記憶装置
7 補助記憶装置
8 出力装置
9,11 プログラム
10,12 データ
13 知識ベースプログラム
14 リレーショナルデータベース
60,80 ノード
61,118 属性
62,72 物と物をつなぐ矢印
63 物と属性をつなぐ矢印
70 物である神経細胞
71 属性である神経細胞
81,88 物識別子
82 順方向ポインタ
83 逆方向ポインタ
84 知識ネットワーク情報
85,102 知識ネットワーク識別子
86 物の名前と物識別子との対応表
87 物の名前
100 重み
101 知識ネットワークと属性との対応表
103 属性識別子
110 感覚器官
111 物の領域
112 記憶領域
114,1020 特徴抽出部
115 属性領域
119 特徴データ
120 物
121 三角形
122 点
123 線
124 面
126 目
127 鼻
128 口
129 駒を置く場所
152 送信装置
153 送信部
155 通信路
157 受信部
158 通信装置識別子
159 受信可能識別子
160 制御部
161 送信受付部
162 検索部
163 アドレス
164 識別データ
165,166,167,168 知識の集まり
169,170,171 知識ネットワーク
1000 知識ベースシステム
1010 データ入力部
1020 特徴抽出部
1030 データ比較部
1040 データ格納部
1050 学習部
1060 検索部
1070 演算部
1080 出力部
Claims (25)
- 知識ベースシステムであって、
知識ベースを記憶している記憶部と、
前記記憶部に記憶された知識ベースに対して論理演算を行う演算部とを備え、
前記知識ベースは、物を識別する物識別子と、前記物がもつ少なくとも一つの属性であって、当該物の物識別子と対応づけられた属性とを含み、
前記属性は、当該属性を識別する属性識別子と、当該属性を表す少なくとも一つのデータであって、当該属性を識別する属性識別子に対応づけられた特徴データ、及び当該属性を表す言葉に対応付けられたデータであって、当該属性識別子に対応付けられた識別データのうちの少なくとも一方とを含み、
前記物識別子は、物を表す言葉ではなく、かつ、それ自体で意味を持たない記号で構成され、
前記属性識別子は、属性を表す言葉ではなく、かつ、それ自体で意味を持たない記号で構成される
知識ベースシステム。 - 前記特徴データは、当該特徴データに対応づけられた属性識別子が識別する属性の形、音、香、味、色、圧力、温度、長さ、座標値、及び面積の少なくとも一つを表すデータである
請求項1記載の知識ベースシステム。 - 前記知識ベースシステムはさらに、
前記属性に関する情報を取得する入力部と、
前記入力部で取得された情報から、前記特徴データ及び前記識別データの少なくとも一つを抽出する特徴抽出部と、
前記特徴抽出部で抽出された前記特徴データ及び前記識別データの少なくとも一つを、前記属性を識別する属性識別子と対応づけて前記知識ベースに格納するデータ格納部とを備える
請求項1記載の知識ベースシステム。 - 前記属性識別子は、主識別子と副識別子とを含み、
前記属性識別子は、前記主識別子と前記副識別子との組み合わせによって、前記属性を識別する
請求項1記載の知識ベースシステム。 - 前記演算部は、少なくとも、一つの物が演算の対象として指定されると指定された当該物がもつ属性を対象として前記論理演算を行う
請求項1記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、
個々の前記属性に、当該属性が真であるか偽であるかを示す真偽値が対応付けられている場合、同じ属性識別子を持ち、かつ、対応づけられた真偽値が共に真であるか又は共に偽である属性だけを集め、
個々の属性に前記真偽値が対応付けられていない場合、同じ属性識別子を持つ属性だけを集めることで、新たな属性の集合を生成するAND演算を行う
請求項5記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、少なくともいずれかの属性の集合に属する属性を集めることで、新たな属性の集合を生成するOR演算を行う
請求項5記載の知識ベースシステム。 - 個々の前記属性には、当該属性が真であるか偽であるかを示す真偽値が対応付けられており、
前記演算部はさらに、少なくとも一つの属性からなる属性の集合が1つ指定された場合において、当該属性の集合に含まれる各属性について、対応づけられた真偽値が真の場合は偽に変えた属性を生成し、真偽値が偽の場合は真に変えた属性を生成することで、真偽値を変更した新たな属性の集合を生成するNOT演算を行う
請求項5記載の知識ベースシステム。 - 個々の前記属性には、当該属性が真であるか偽であるかを示す真偽値が対応付けられており、
前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、同じ属性識別子を持ち、かつ、対応づけられた真偽値が共に真であるか又は共に偽である属性だけを集めることで、新たな属性の集合を生成し、生成した新たな属性の集合に含まれる各属性について、対応づけられた真偽値が真の場合は偽に変えた属性を生成し、真偽値が偽の場合は真に変えた属性を生成することで、真偽値を変更した更に新たな属性の集合を生成するNAND演算を行う
請求項5記載の知識ベースシステム。 - 個々の前記属性には、当該属性が真であるか偽であるかを示す真偽値が対応付けられており、
前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、少なくともいずれかの属性の集合に属する属性を集めることで、新たな属性の集合を生成し、生成した新たな属性の集合に含まれる各属性について、対応づけられた真偽値が真の場合は偽に変えた属性を生成し、真偽値が偽の場合は真に変えた属性を生成することで、真偽値を変更した更に新たな属性の集合を生成するNOR演算を行う
請求項5記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、
個々の前記属性に、当該属性が真であるか偽であるかを示す真偽値が対応づけられている場合、いずれの属性の集合にも含まれ、かつ、対応づけられた真偽値が同じである属性の数を共通度として計数し、
個々の前記属性に前記真偽値が対応づけられていない場合、いずれの属性の集合にも含まれる属性の数を共通度として計数する
請求項5記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、
個々の前記属性に、当該属性が真であるか偽であるかを示す真偽値が対応づけられている場合、いずれの属性の集合にも含まれ、かつ、対応づけられた真偽値が異なる属性の数を非共通度として計数し、
個々の属性に前記真偽値が対応づけられていない場合、いずれかの属性の集合にだけ含まれる属性の数を非共通度として計数する
請求項5記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、
個々の前記属性に、当該属性が真であるか偽であるかを示す真偽値が対応づけられている場合、主識別子が同じで、かつ、対応づけられた真偽値が同じである属性の数を類似度として計数し、
個々の前記属性に前記真偽値が対応づけられていない場合、主識別子が同じである属性の数を類似度として計数する
請求項5記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、
個々の前記属性に、当該属性が真であるか偽であるかを示す真偽値が対応づけられている場合、主識別子が同じで、かつ、対応づけられた真偽値が異なる属性の数を非類似度として計数し、
個々の前記属性に前記真偽値が対応づけられていない場合、いずれかの属性の集合だけに含まれる属性の数を非類似度として計数する
請求項5記載の知識ベースシステム。 - 個々の前記属性には、当該属性が真であるか偽であるかを示す真偽値が対応付けられており、
前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、少なくともいずれかの属性の集合に属する属性から、いずれの属性の集合にも含まれ、かつ、対応づけられた真偽値が異なる属性を除くことで、新たな属性の集合を生成するOR演算を行う
請求項5記載の知識ベースシステム。 - 前記演算部はさらに、少なくとも一つの属性からなる属性の集合が2つ指定された場合において、いずれの属性の集合にも属する属性を除くことで、新たな属性の集合を生成するXOR演算を行う
請求項5記載の知識ベースシステム。 - 前記演算部は、
少なくとも一つの属性が指定されると、指定された属性に対応づけられた物識別子を集めることで、当該属性をもつ知識の集まりを生成し、
物が指定されると、当該物の物識別子が前記知識の集まりに属するか否かを判断する
請求項1記載の知識ベースシステム。 - 前記演算部はさらに、
生成した前記知識の集まりを識別する物識別子と、
前記指定された少なくとも一つの属性を識別する属性識別子、及び抽象的な物であることを示す属性の属性識別子とを対応付けることで、抽象的な物を新たに生成する
請求項17に記載の知識ベースシステム。 - 前記知識ベースはさらに、物に対応づけられたノードのつながりである知識ネットワークを含み、
前記ノードは、当該ノードに対応する物識別子と、当該ノードが属する知識ネットワークに関する情報である知識ネットワーク情報とを含み、
前記知識ネットワーク情報は、前記知識ネットワークを識別する知識ネットワーク識別子と、前記知識ネットワークにおける当該ノードとつながる他のノードへのポインタとを含む
請求項1記載の知識ベースシステム。 - 前記演算部は、指定された知識ネットワーク識別子で識別された知識ネットワークにおいて、当該知識ネットワークを構成するノードに含まれる物識別子と知識ネットワーク情報とを参照することで、指定された物とつながっている他の物を特定する
請求項19記載の知識ベースシステム。 - 前記演算部は、属性識別子が指定されると、当該属性識別子に対応づけられた知識ネットワークを検索し、検索された知識ネットワークの知識ネットワーク識別子を取得する
請求項19記載の知識ベースシステム。 - 前記演算部はさらに、
前記知識ネットワークを識別する物識別子と、
前記知識ネットワークに対応付けられた少なくとも1つの属性識別子、及び抽象的な物であることを示す属性の属性識別子とを対応付けることで、抽象的な物を新たに生成する
請求項21に記載の知識ベースシステム。 - 知識ベースシステムにおける論理演算方法であって、
請求項1記載の知識ベースシステムが備える知識ベースに対して、少なくとも請求項1~22のいずれか1項に記載の演算部による処理を行うステップを含む
論理演算方法。 - 知識ベースシステムのためのプログラムであって、
請求項23記載の論理演算方法に含まれるステップをコンピュータに実行させる
プログラム。 - 知識ベースシステムのためのコンピュータ読み取り可能な記録媒体であって、
請求項1記載の知識ベースと、
請求項24記載のプログラムと
が記録された記録媒体。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2010541357A JP4829381B2 (ja) | 2009-05-18 | 2010-05-18 | 知識ベースシステム、論理演算方法、プログラム、及び記録媒体 |
US13/320,801 US8818930B2 (en) | 2009-05-18 | 2010-05-18 | Knowledge base system, logic operation method, program, and recording medium |
EP20100777557 EP2472447A4 (en) | 2009-05-18 | 2010-05-18 | BASIC KNOWLEDGE SYSTEM, LOGICAL OPERATION METHOD, PROGRAM, AND STORAGE MEDIUM |
Applications Claiming Priority (8)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2009-119693 | 2009-05-18 | ||
JP2009119693 | 2009-05-18 | ||
JP2009211424 | 2009-09-14 | ||
JP2009-211424 | 2009-09-14 | ||
JP2009-273583 | 2009-12-01 | ||
JP2009273583 | 2009-12-01 | ||
JP2010-009960 | 2010-01-20 | ||
JP2010009960 | 2010-01-20 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2010134319A1 true WO2010134319A1 (ja) | 2010-11-25 |
Family
ID=43126012
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2010/003327 WO2010134319A1 (ja) | 2009-05-18 | 2010-05-18 | 知識ベースシステム、論理演算方法、プログラム、及び記録媒体 |
Country Status (4)
Country | Link |
---|---|
US (1) | US8818930B2 (ja) |
EP (1) | EP2472447A4 (ja) |
JP (4) | JP4829381B2 (ja) |
WO (1) | WO2010134319A1 (ja) |
Families Citing this family (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP4829381B2 (ja) * | 2009-05-18 | 2011-12-07 | 隆敏 柳瀬 | 知識ベースシステム、論理演算方法、プログラム、及び記録媒体 |
WO2012002215A1 (ja) * | 2010-07-02 | 2012-01-05 | Yanase Takatoshi | 論理演算システム |
US9690457B2 (en) * | 2012-08-24 | 2017-06-27 | Empire Technology Development Llc | Virtual reality applications |
US9607436B2 (en) | 2012-08-27 | 2017-03-28 | Empire Technology Development Llc | Generating augmented reality exemplars |
US10223410B2 (en) * | 2014-01-06 | 2019-03-05 | Cisco Technology, Inc. | Method and system for acquisition, normalization, matching, and enrichment of data |
US10867527B2 (en) * | 2014-09-01 | 2020-12-15 | 5Lion Horus Tech Llc. | Process and wearable device equipped with stereoscopic vision for helping the user |
US11468340B2 (en) * | 2015-02-09 | 2022-10-11 | Xerox Corporation | Method and system for using data storage technology as a repository of functional logic knowledge for use in a knowledge abstraction layer subsystem |
EP3545372B1 (en) | 2016-11-23 | 2021-12-29 | Carrier Corporation | Building management system having knowledge base |
US10972306B2 (en) | 2016-11-23 | 2021-04-06 | Carrier Corporation | Building management system having event reporting |
KR101890207B1 (ko) * | 2017-01-03 | 2018-08-22 | 네이버 주식회사 | 개체명 연결 방법, 장치 및 컴퓨터 프로그램 |
CN107391682B (zh) * | 2017-07-24 | 2020-06-09 | 京东方科技集团股份有限公司 | 知识验证方法、知识验证设备以及存储介质 |
JP6814482B2 (ja) * | 2017-11-29 | 2021-01-20 | 株式会社医療情報技術研究所 | 知識管理システム |
CN108052612A (zh) * | 2017-12-14 | 2018-05-18 | 马文波 | 大数据的list显示方法和装置 |
DE102018100895A1 (de) * | 2018-01-16 | 2019-07-18 | Zoe Life Technologies Holding AG | Währungseinheiten für Wissen |
JP6913308B2 (ja) * | 2018-07-04 | 2021-08-04 | 株式会社医療情報技術研究所 | 医療文書管理システム |
US20200057959A1 (en) | 2018-08-15 | 2020-02-20 | Salesforce.Com, Inc. | Reducing instances of inclusion of data associated with hindsight bias in a training set of data for a machine learning system |
US11263400B2 (en) * | 2019-07-05 | 2022-03-01 | Google Llc | Identifying entity attribute relations |
CN110674224B (zh) * | 2019-08-02 | 2022-05-27 | 北京百度网讯科技有限公司 | 实体数据的处理方法、装置、设备及计算机可读存储介质 |
CN113360496A (zh) * | 2021-05-26 | 2021-09-07 | 国网能源研究院有限公司 | 一种构建元数据标签库的方法及装置 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0276033A (ja) * | 1988-09-13 | 1990-03-15 | Fujitsu Ltd | 知識整理支援装置 |
JPH10275085A (ja) * | 1997-03-31 | 1998-10-13 | Toshiba Corp | 概念階層作成装置及び概念階層作成方法 |
JPH1166085A (ja) * | 1997-08-19 | 1999-03-09 | Matsushita Electric Ind Co Ltd | 情報送信装置および受信装置 |
JP2002197114A (ja) * | 2000-12-27 | 2002-07-12 | Beacon Information Technology:Kk | データベース管理システム、顧客管理システム、記録媒体 |
JP2005196382A (ja) * | 2004-01-06 | 2005-07-21 | Nippon Telegr & Teleph Corp <Ntt> | 情報管理システムによるデータ登録方法、データ参照方法、ならびに、情報管理システムにおけるマスタデータベースサーバ |
JP2006514606A (ja) | 2002-07-12 | 2006-05-11 | シーエスエル、リミテッド | 疎水性タンパク質の発現 |
Family Cites Families (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2065473C (en) * | 1991-04-25 | 1999-08-10 | Kouichi Shibao | Knowledge base management system for an information reasoning apparatus |
JPH0594478A (ja) | 1991-10-02 | 1993-04-16 | Matsushita Electric Ind Co Ltd | 画像データベースシステム |
GB2278467B (en) * | 1993-05-20 | 1998-04-01 | Canon Res Ct Europe Ltd | Apparatus and method for data processing and/or for control |
US5546526A (en) * | 1993-12-10 | 1996-08-13 | International Business Machines Corporation | Reconfiguration of database by interactive manipulation of icons |
JPH07334368A (ja) * | 1994-06-08 | 1995-12-22 | Hitachi Ltd | 知識ベースシステムおよび認識システム |
US5701400A (en) * | 1995-03-08 | 1997-12-23 | Amado; Carlos Armando | Method and apparatus for applying if-then-else rules to data sets in a relational data base and generating from the results of application of said rules a database of diagnostics linked to said data sets to aid executive analysis of financial data |
JPH08305708A (ja) | 1995-04-28 | 1996-11-22 | Nippon Telegr & Teleph Corp <Ntt> | 説明文生成方法及び装置 |
US6314570B1 (en) * | 1996-02-08 | 2001-11-06 | Matsushita Electric Industrial Co., Ltd. | Data processing apparatus for facilitating data selection and data processing in at television environment with reusable menu structures |
US6353813B1 (en) * | 1998-01-22 | 2002-03-05 | Microsoft Corporation | Method and apparatus, using attribute set harmonization and default attribute values, for matching entities and predicting an attribute of an entity |
US6499024B1 (en) * | 1999-08-24 | 2002-12-24 | Stream International, Inc. | Method and system for development of a knowledge base system |
US6493697B1 (en) * | 1999-08-24 | 2002-12-10 | Stream International, Inc. | Method of selecting desired domains and for developing a seeding methodology for a knowledge base system |
US6560589B1 (en) * | 1999-08-24 | 2003-05-06 | Stream International, Inc. | Method and system for use and maintenance of a knowledge base system |
US6591258B1 (en) * | 1999-08-24 | 2003-07-08 | Stream International, Inc. | Method of incorporating knowledge into a knowledge base system |
JP2001229382A (ja) | 2000-02-17 | 2001-08-24 | Nippon Telegr & Teleph Corp <Ntt> | 情報蓄積装置,情報検索装置および情報蓄積方法,情報検索方法並びにこれらの方法を記録した記録媒体 |
AU2001264611A1 (en) * | 2000-05-17 | 2001-11-26 | Microsoft Corporation | System and method for matching a textual input to a lexical knowledge base and for utilizing results of that match |
JP2002041540A (ja) * | 2000-07-28 | 2002-02-08 | Shinichiro Okude | 連想・推論機能を備えた検索システムおよび、それに用いられる、記録媒体献金 |
US20020169779A1 (en) * | 2001-05-10 | 2002-11-14 | Shian-Shyong Tseng | Object oriented knowledge base management system |
US20030105880A1 (en) * | 2001-12-04 | 2003-06-05 | Koninklijke Philips Electronics N.V. | Distributed processing, storage, and transmision of multimedia information |
US20040122705A1 (en) * | 2002-12-18 | 2004-06-24 | Sabol John M. | Multilevel integrated medical knowledge base system and method |
US20050154535A1 (en) * | 2004-01-09 | 2005-07-14 | Genstruct, Inc. | Method, system and apparatus for assembling and using biological knowledge |
JP2005050315A (ja) | 2004-06-28 | 2005-02-24 | Nec Saitama Ltd | 携帯端末 |
WO2007000893A1 (ja) | 2005-06-14 | 2007-01-04 | Mitsuhiro Tanada | 情報空間処理装置、プログラム、および、方法 |
US8666928B2 (en) * | 2005-08-01 | 2014-03-04 | Evi Technologies Limited | Knowledge repository |
JP2007188469A (ja) | 2005-12-13 | 2007-07-26 | Mitsuhiro Tanada | 情報空間処理装置、プログラム、および、方法 |
JP2009075714A (ja) | 2007-09-19 | 2009-04-09 | Sekisui House Ltd | 情報検索システム |
JP5094478B2 (ja) | 2008-03-06 | 2012-12-12 | 株式会社日立製作所 | 受信装置 |
JP4829381B2 (ja) * | 2009-05-18 | 2011-12-07 | 隆敏 柳瀬 | 知識ベースシステム、論理演算方法、プログラム、及び記録媒体 |
WO2012002215A1 (ja) * | 2010-07-02 | 2012-01-05 | Yanase Takatoshi | 論理演算システム |
-
2010
- 2010-05-18 JP JP2010541357A patent/JP4829381B2/ja not_active Expired - Fee Related
- 2010-05-18 WO PCT/JP2010/003327 patent/WO2010134319A1/ja active Application Filing
- 2010-05-18 EP EP20100777557 patent/EP2472447A4/en not_active Withdrawn
- 2010-05-18 US US13/320,801 patent/US8818930B2/en active Active
-
2011
- 2011-06-15 JP JP2011132779A patent/JP4865925B2/ja not_active Expired - Fee Related
- 2011-09-16 JP JP2011202849A patent/JP4891460B2/ja active Active
- 2011-12-01 JP JP2011263928A patent/JP2012079331A/ja active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0276033A (ja) * | 1988-09-13 | 1990-03-15 | Fujitsu Ltd | 知識整理支援装置 |
JPH10275085A (ja) * | 1997-03-31 | 1998-10-13 | Toshiba Corp | 概念階層作成装置及び概念階層作成方法 |
JPH1166085A (ja) * | 1997-08-19 | 1999-03-09 | Matsushita Electric Ind Co Ltd | 情報送信装置および受信装置 |
JP2002197114A (ja) * | 2000-12-27 | 2002-07-12 | Beacon Information Technology:Kk | データベース管理システム、顧客管理システム、記録媒体 |
JP2006514606A (ja) | 2002-07-12 | 2006-05-11 | シーエスエル、リミテッド | 疎水性タンパク質の発現 |
JP2005196382A (ja) * | 2004-01-06 | 2005-07-21 | Nippon Telegr & Teleph Corp <Ntt> | 情報管理システムによるデータ登録方法、データ参照方法、ならびに、情報管理システムにおけるマスタデータベースサーバ |
Non-Patent Citations (1)
Title |
---|
See also references of EP2472447A4 |
Also Published As
Publication number | Publication date |
---|---|
JP2011222037A (ja) | 2011-11-04 |
EP2472447A1 (en) | 2012-07-04 |
JP4891460B2 (ja) | 2012-03-07 |
JP4829381B2 (ja) | 2011-12-07 |
JP4865925B2 (ja) | 2012-02-01 |
US20120072387A1 (en) | 2012-03-22 |
EP2472447A4 (en) | 2014-08-20 |
US8818930B2 (en) | 2014-08-26 |
JP2012033181A (ja) | 2012-02-16 |
JPWO2010134319A1 (ja) | 2012-11-08 |
JP2012079331A (ja) | 2012-04-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4891460B2 (ja) | 知識ベースシステム、論理演算方法、プログラム、及び記録媒体 | |
US11868883B1 (en) | Intelligent control with hierarchical stacked neural networks | |
González García et al. | A review of artificial intelligence in the internet of things | |
Molnar | Interpretable machine learning | |
US11514305B1 (en) | Intelligent control with hierarchical stacked neural networks | |
Hinkel et al. | Enhancing the Ostrom social-ecological system framework through formalization | |
Kaklauskas | Biometric and intelligent decision making support | |
Domingos | Prospects and challenges for multi-relational data mining | |
Fan et al. | Attribute-oriented cognitive concept learning strategy: a multi-level method | |
Sosnovshchenko et al. | Machine Learning with Swift: Artificial Intelligence for IOS | |
CN113516164A (zh) | 一种融合知识图谱与深度学习的果树病虫害诊断方法 | |
Amoore | What does it mean to govern with algorithms | |
Yao | Design and simulation of integrated education information teaching system based on fuzzy logic | |
Klir | Systems science: a guided tour | |
Kelley et al. | A cognitive robotics system: the symbolic and sub-symbolic robotic intelligence control system (SS-RICS) | |
Woof et al. | A framework for end-to-end learning on semantic tree-structured data | |
Tresp et al. | The tensor brain: A unified theory of perception, memory, and semantic decoding | |
Li | Application of data mining methods in the study of crime based on international data sources | |
Srinivasan et al. | Quantifying Confounding Bias in Generative Art: A Case Study | |
Barceló | Computational intelligence in archaeology. State of the art | |
Catarau-Cotutiu et al. | AIGenC: An AI generalisation model via creativity | |
Gunasinghe et al. | A biologically inspired neural clustering model for capturing patterns from incomplete data | |
Hasan et al. | Toward enhancement of deep learning techniques using fuzzy logic: a survey | |
Mukherjee | Integrating common sense knowledge in visual question answering systems | |
Singh et al. | Answering Questions in Natural Language About Images Using Deep Learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2010541357 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10777557 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13320801 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2010777557 Country of ref document: EP |